LLM Registry API: schemas

The request and response bodies of the operations in this API.

AgentAuthoringResponseDto

Property	Type	Required	Description
`abilities`	array of string	yes
`capability`	AgentCapabilityLayerDto or null	no
`description`	string or null	no
`displayName`	string	yes
`draftUpdatedAt`	string (date-time)	yes	Mutable draft concurrency token.
`durableWorkspace`	DurableWorkspaceRefDto or null	no
`id`	string	yes	Stable Agent identity id.
`internetEgressCeiling`	WorkerInternetEgress	no	Ceiling on whether the worker pod running this agent may open sockets to the public internet. A ceiling the agent ACCEPTS, never a grant: the effective mode is the most restrictive of the deployment, the owning Space, this value and the session, so `open` here widens nothing. What THIS agent itself states — NOT joined with the agent it `extends`; the resolved-agent and agent-blocks responses carry the joined value. ABSENT when it declares none — there is no default.
`liveRevision`	AgentLiveRevisionDto or null	no
`reachTier`	AgentReachTier	yes
`root`	AgentNodeDto	yes
`scope`	ScopeSource	yes
`sessionLifecycle`	AgentSessionLifecycle or null	no
`slug`	string	yes
`standingSession`	StandingSessionRefDto or null	no
`state`	one of `"draft"`, `"live"`	yes
`workspace`	AgentWorkspaceDto or null	no
`workspaceSharing`	WorkspaceSharingPolicy	yes

AgentAuthoringResponseDtoDataArrayEnvelope

Property	Type	Required	Description
`data`	array of AgentAuthoringResponseDto	yes

AgentAuthoringResponseDtoDataEnvelope

Property	Type	Required	Description
`data`	AgentAuthoringResponseDto	yes

AgentBiomeAvailabilityDto

Property	Type	Required	Description
`exclude`	array of string	no	Deny-list of biome ids. A contribution whose owning biome is listed is always dropped, even if it also appears in include. Absent/empty ⇒ nothing force-excluded.
`include`	array of string	no	Allow-list of biome ids. When non-empty, ONLY contributions whose owning biome is listed survive (non-biome contributions always pass). Absent/empty ⇒ all biomes allowed unless excluded.

AgentCapabilityAnalysisDto

Property	Type	Required	Description
`reachTier`	AgentReachTier	yes	The agent's reach-tier ceiling — the provisioning input.
`requiredCapabilities`	array of string	yes	Deduped, sorted union of every CapabilityRef the agent needs to drive its armed tools + skills across the whole sub-agent tree.
`slug`	string	yes
`unresolvedSkillSlugs`	array of string	yes	Armed skill slugs with no capability data available — see the skill-registry surfacing gap in capability-derivation.ts.
`unresolvedTools`	array of UnresolvedToolDto	yes	Armed tool refs not resolvable to capabilities from the canonical MCP catalog (flagged, never silently dropped).
`version`	string	yes

AgentCapabilityAnalysisDtoDataEnvelope

Property	Type	Required	Description
`data`	AgentCapabilityAnalysisDto	yes

AgentCapabilityApertureDto

Property	Type	Required
`defaultEffect`	one of `"read"`, `"write"`, `"admin"` or null	no
`domains`	map of one of `"read"`, `"write"`, `"admin"`	no
`kind`	one of `"unrestricted"`, `"armed"`, `"scoped"`, `"representative"`	yes
`refs`	array of string	no

AgentCapabilityLayerDto

Property	Type	Required	Description
`aperture`	AgentCapabilityApertureDto	yes	Explicit, closed aperture union; never inferred from omission.
`mayEditRepos`	one of `"never"`, `"when-granted"`	no
`mayWriteKnowledgeBase`	boolean	no
`writePolicy`	one of `"read-only"`, `"own-pages-only"`, `"full-write"`	no

AgentCapabilitySetExpressionDto

Property	Type	Required	Description
`exclude`	array of string	yes
`include`	array of string	yes

AgentCredentialDto

Property	Type	Required	Description
`kind`	CredentialKind	yes
`name`	string	yes
`required`	boolean	yes	When true the run fails fast if the credential is unresolvable.
`sourceRef`	string	yes	Opaque source ref the credential broker resolves the secret from.

AgentDraftRingEntryDto

Property	Type	Required	Description
`baseRevisionId`	string or null	yes
`contentHash`	string	yes	Canonical content hash. Two entries never share one — the ring only appends when the hash changes.
`createdAt`	string (date-time)	yes
`id`	string	yes
`isCurrent`	boolean	yes	True for the entry the Agent identity currently points at.
`restoredFromRevisionId`	string or null	yes
`updatedAt`	string (date-time)	yes
`updatedBy`	string	yes

AgentDraftRingEntryDtoDataArrayEnvelope

Property	Type	Required	Description
`data`	array of AgentDraftRingEntryDto	yes

AgentEnvVarDto

Property	Type	Required	Description
`name`	string	yes
`value`	string	yes	Authored value. May contain `${input.x}` interpolation — the resolution service resolves it before projecting onto WorkspaceSpec.

AgentExecutionMode

Type: one of "primary", "subagent"

AgentKernelResponseDto

Property	Type	Required	Description
`avatarSeed`	string or null	no
`avatarStyle`	string	yes
`biomeId`	string or null	no	Biome id when source='biome', otherwise null.
`capability`	AgentCapabilityLayerDto	yes	The agent's resolved capability envelope and explicit aperture. Always present.
`createdAt`	string (date-time)	yes
`createdBy`	string	yes
`definitionHash`	object or null	no
`description`	string or null	no
`displayName`	string	yes
`executionMode`	AgentExecutionMode	yes
`forkedFrom`	string or null	no	Origin reference (`<source>:<biomeId\|""">:<slug>`) when this row was forked from a biome/system row. Survives uninstall — UI surfaces "fork without origin" if the origin is gone.
`hidden`	boolean	yes	Hidden from user-facing pickers. Stays invocable via `task` delegation and workflow `agentRef:`.
`id`	string	yes
`identityVersion`	number	yes
`isActive`	boolean	yes
`mcpServices`	array of string	yes
`modelLane`	string or null	yes	The model LANE this agent routes on — rules select on it. An OPEN vocabulary: every key is declared by a `model-lane` contribution, and `GET /work-kinds?kind=model-lane` enumerates what this organisation may use. Distinct from AgentRunRole (permissions).
`orgId`	string or null	yes	The owning organisation, or null for a platform-owned agent (`system` / `biome` tier).
`ownerUserId`	string or null	no	Owning user id when source='user', otherwise null.
`projectId`	string or null	no
`revisionNumber`	number	yes	Server-generated immutable revision number.
`skills`	array of string	yes
`slug`	string	yes
`source`	ScopeSource	yes	5-tier owner-scope label for UI source-filter and fork CTA. Matches the `source` column on the row; defaults to `org` for legacy rows.
`supportsAgentSessions`	boolean	yes
`systemPrompt`	string	yes
`temperature`	number or null	no	`null` defers to the engine/model default.
`updatedAt`	string (date-time)	yes
`workspace`	AgentWorkspaceDto or null	no	Authored workspace intent for launch compatibility checks.

AgentKernelResponseDtoDataArrayEnvelope

Property	Type	Required	Description
`data`	array of AgentKernelResponseDto	yes

AgentKernelResponseDtoDataEnvelope

Property	Type	Required	Description
`data`	AgentKernelResponseDto	yes

AgentKernelResponseDtoPaginatedEnvelope

Property	Type	Required	Description
`data`	array of AgentKernelResponseDto	yes
`pagination`	PaginationMeta	yes

AgentLiveRevisionDto

Property	Type	Required
`contentHash`	string	yes
`createdAt`	string (date-time)	yes
`id`	string	yes
`number`	number	yes

AgentModelOverrideDto

Property	Type	Required	Description
`kind`	one of `"concrete"`, `"strategy"`	yes
`modelId`	string	no
`modelLane`	string	no	Work kind this node's model resolves through (kind=strategy). OPEN vocabulary with no kernel core: every lane is declared by a `model-lane` contribution, so `GET /work-kinds?kind=model-lane` enumerates the lanes this organisation may use. Checked against the org's visible set on write.
`providerSlug`	string	no
`temperature`	number	no

AgentModelOverrideResponseDto

Property	Type	Required	Description
`agentSlug`	string	yes
`createdAt`	string	yes
`id`	string	yes
`model`	object	no	Resolved model info (included when fetching).
`modelId`	string	yes
`orgId`	string	yes
`scope`	one of `"org"`, `"user"`, `"project"`	yes
`scopeId`	string	yes
`temperature`	number or null	no
`updatedAt`	string	yes

AgentModelOverrideResponseDtoDataArrayEnvelope

Property	Type	Required	Description
`data`	array of AgentModelOverrideResponseDto	yes

AgentModelOverrideResponseDtoDataEnvelope

Property	Type	Required	Description
`data`	AgentModelOverrideResponseDto	yes

AgentMountLayoutDto

Property	Type	Required	Description
`agentRunConfig`	AgentRunConfigDto	yes
`inputs`	object	yes	Raw spec.inputs block — manifest DSL ManifestInputDeclaration map.
`mounts`	object	yes	Raw spec.mounts block — manifest DSL ManifestMountDeclaration map keyed by AWP slot key.
`seedFiles`	array of object	yes	Raw spec.seedFiles array — manifest DSL ManifestSeedFile entries.

AgentNodeDto

Property	Type	Required	Description
`agentContentHash`	string	no	Server-stamped, resolve-only: SHA-256 of the STABLE revision of the agent this node references. Rejected on write.
`alias`	string	no	Alias when the same agent appears more than once.
`biomeAvailability`	AgentBiomeAvailabilityDto	no	Per-node biome opt-in/opt-out (DBM.2). A POST-scope filter the resolver applies over this node's biome-contributed references — its child agents (and, downstream, skills/tools). `exclude` wins; a non-empty `include` is an allow-list; absent ⇒ no filtering (implicit-by-scope). Mirrors the kernel `BiomeAvailabilityScoped` mixin on `AgentNode`.
`children`	array of AgentNodeDto	yes	Sub-agents — recursive composition nodes.
`delegationBoundary`	AgentCapabilitySetExpressionDto	no	DELEGATION BOUNDARY — a ceiling on the authority this node may be granted by the agent that delegates to it (kernel `AgentNode.delegationBoundary`). `effective(this) = effective(parent) ∩ own(this) ∩ delegationBoundary`; absent ⇒ no ADDITIONAL attenuation — the parent's authority still bounds this node. Meaningless on the ROOT node (nothing delegates to it) and refused there by `AgentNodeValidator`.
`instructions`	string	no	Per-node instructions appended after the referenced agent definition's systemPrompt. Extends (does not replace) the agent prompt for this node only — sibling and child nodes are unaffected. Length is the author’s decision — no product cap; coarse abuse bounding is handled by the HTTP body-size limit.
`kernel`	AgentRefDto	yes
`modelOverride`	AgentModelOverrideDto	no
`permission`	object	no	Per-node permission override map (action or per-target map per key), deep-merged over the referenced agent definition permission. e.g. { edit: "deny", webfetch: "deny", task: { "*": "deny", "kb-researcher": "allow" } }
`promptMode`	one of `"append"`, `"replace"`	no	How this node’s instructions combine with the referenced agent systemPrompt. "append" (default) appends after the base prompt; "replace" makes instructions the prompt body (config still inherited).
`skills`	array of AgentSkillRefDto	yes
`tools`	array of AgentToolRefDto	yes

AgentOutputSurfaceDto

Property	Type	Required	Description
`autoOpen`	boolean	no	Auto-open the output surface pane on first ready signal.
`defaultDocument`	string	no	Default document name served for directory requests when `kind === 'static'`. Defaults to `index.html`.
`healthPath`	string	no
`kind`	OutputSurfaceKind	yes
`mode`	one of `"single"`, `"multi"`	no
`port`	number	no
`root`	string	no	Workspace-relative directory served by the baked static-artifact-server. Required when `kind === 'static'`, rejected for other kinds.

AgentOwnerSpaceDto

Property	Type	Required
`appId`	string	no
`biomeId`	string	no
`orgId`	string	no
`projectId`	string	no
`sessionId`	string	no
`tier`	one of `"system"`, `"org"`, `"project"`, `"app"`, `"session"`, `"biome"`, `"user"`	yes
`userId`	string	no

AgentOwnerSpaceRefDto

Property	Type	Required
`fence`	one of `"global"`, `"tenant"`	yes
`orgId`	string or null	yes
`space`	AgentOwnerSpaceDto	yes

AgentPermissionsDto

Property	Type	Required	Description
`tools`	AgentPermissionsToolsDto	yes

AgentPermissionsToolsDto

Property	Type	Required	Description
`allow`	array of ToolsetKey	yes
`deny`	array of ToolsetKey	yes

AgentPersistenceDto

Property	Type	Required	Description
`overrides`	object	no	Per-slot persistence-policy overrides keyed by AWP slot key. Values are the SlotPersistencePolicy union; only git-push slots are overridable.
`paths`	array of string	yes	Workspace-relative paths. No leading "/", no "..".

AgentProfileDto

Property	Type	Required	Description
`avatarSeed`	object or null	no
`avatarStyle`	string	yes
`biomeId`	string or null	no	Biome id when source='biome', otherwise null. Used by the FE talent-pool view to group biome-contributed agents under a single category bucket.
`description`	object or null	no
`displayName`	string	yes
`executionMode`	AgentExecutionMode	yes	Role hint resolved from the agent's source manifest `mode:` frontmatter. `primary` agents lead a session; `subagent` agents are designed to be invoked via a primary's `task` tool. The distinction is a UI hint — any agent may be bound in either role at session/step time.
`id`	string	yes
`identityVersion`	number	yes
`isActive`	boolean	yes
`modelLane`	string or null	yes	The model LANE this agent routes on — rules select on it. An OPEN vocabulary: every key is declared by a `model-lane` contribution, and `GET /work-kinds?kind=model-lane` enumerates what this organisation may use. Distinct from AgentRunRole (permissions).
`skills`	array of string	yes
`slug`	string	yes
`source`	string	yes	Scope tier the row lives in. `system` rows ship with the platform; `biome` rows are contributed by an installed biome and grouped by `biomeId`; `org` rows are authored in the org; `project` rows are scoped to a single project.

AgentProfileDtoDataEnvelope

Property	Type	Required	Description
`data`	AgentProfileDto	yes

AgentProfileDtoPaginatedEnvelope

Property	Type	Required	Description
`data`	array of AgentProfileDto	yes
`pagination`	PaginationMeta	yes

AgentReachTier

The root composition agent's reach-tier permission ceiling (Chunk 2). A composition row with no explicit tier resolves to as-owner (the default), so this field is always populated. agent-session-api persists it onto the session and stamps it onto ExecutionContext.reachTier so authorization-api applies the tier constraint inside the verdict.

Type: one of "sandbox", "as-owner", "connected", "full"

AgentRefDto

Property	Type	Required	Description
`slug`	string	yes
`version`	string	no	Pinned version; omitted = latest.

AgentRevisionHistoryEntryDto

Property	Type	Required	Description
`contentHash`	string	yes
`createdAt`	string (date-time)	yes
`createdBy`	string	yes
`id`	string	yes
`number`	number	yes
`restoredFromRevisionId`	string or null	no
`sourceProvenance`	object or null	no	Advanced provenance only. May contain sourceVersion/sourceCommit; it is not a version selector.

AgentRevisionHistoryEntryDtoDataArrayEnvelope

Property	Type	Required	Description
`data`	array of AgentRevisionHistoryEntryDto	yes

AgentRollbackResponseDto

Property	Type	Required
`expectedChannelVersion`	number	yes
`identityId`	string	yes
`priorReleaseSetId`	string	yes
`releaseIntentId`	string	yes
`status`	one of `"pending"`, `"reconciling"`, `"applied"`, `"failed"`	yes

AgentRollbackResponseDtoDataEnvelope

Property	Type	Required	Description
`data`	AgentRollbackResponseDto	yes

AgentRunConfigDto

Property	Type	Required	Description
`deliverableSpecRef`	string	no	Deliverable-spec ref. May carry a `${input.x}` token — resolved downstream by the workspace pipeline against the run bind inputs.
`role`	string	yes	Canonical AgentRunRole — drives renderer + system-overlay framing.
`stage`	string	yes	Stage key.

AgentSessionLifecycle

The resolved agent's declared default session lifecycle (ephemeral|persistent), or null when it declares no preference. agent-session-api inherits this as the session lifecyclePolicy unless the launch passes an explicit override; null ⇒ agent-session applies its own ephemeral default last.

Type: one of "ephemeral", "persistent"

AgentSkillRefDto

Property	Type	Required	Description
`slug`	string	yes
`version`	string	no	Pinned version; omitted = latest approved.

AgentTalentPoolStatsDto

Property	Type	Required	Description
`avgDurationMs`	number	yes	Mean wall-clock duration per invocation, in milliseconds. Always 0 while `source` is `unavailable`.
`avgTokensPerInvocation`	number	yes	Mean tokens per invocation. Always 0 while `source` is `unavailable`.
`invocationCount`	number	yes	Recorded invocations. Always 0 while `source` is `unavailable`.
`lastUsedAt`	string (date-time) or null	no	When the agent was last invoked. Always null while `source` is `unavailable`.
`slug`	string	yes	Slug of the agent these stats describe.
`source`	AgentTalentPoolStatsSource	yes	Provenance of the numbers above. Read this before rendering them — `unavailable` means no telemetry exists, not that the agent was never used.

AgentTalentPoolStatsDtoDataEnvelope

Property	Type	Required	Description
`data`	AgentTalentPoolStatsDto	yes

AgentTalentPoolStatsSource

Provenance of the numbers above. Read this before rendering them — unavailable means no telemetry exists, not that the agent was never used.

Type: one of "unavailable"

AgentToolRefDto

Property	Type	Required	Description
`kind`	one of `"provider"`, `"tool"`	yes
`providerKind`	string	yes	Tool provider kind (see ToolProviderKind).
`resourceId`	string	yes
`toolName`	string	no	Canonical MCP tool name. Required when kind = tool.

AgentWorkingFileDto

Property	Type	Required
`format`	one of `"markdown"`, `"html"`, `"json"`, `"yaml"`, `"text"`	yes
`path`	string	yes
`slug`	string	yes
`sourceKind`	string	yes
`sourceRef`	map of string	yes
`syncDirection`	one of `"down-only"`, `"up-only"`, `"bidirectional"`	yes

AgentWorkspaceDto

Property	Type	Required	Description
`credentials`	array of AgentCredentialDto	no	Declared credential requirements projected from the source manifest compiled `spec.credentials[]`. Read by the runtime EnvironmentResolver (snapshot + resume drift detection). Present only when the source manifest authored at least one credential.
`env`	array of AgentEnvVarDto	no
`mountLayout`	AgentMountLayoutDto	no
`outputSurface`	AgentOutputSurfaceDto	no
`permissions`	AgentPermissionsDto	no	Coarse toolset allow/deny gate projected from the source manifest compiled `spec.permissions`. Read by the runtime EnvironmentResolver (snapshot + resume drift detection). Present only when the source manifest authored a non-empty allow or deny list.
`persistence`	AgentPersistenceDto	no
`workingFiles`	array of AgentWorkingFileDto	no	Working files the platform materialises for sessions on this composition. Raw form — `path` / `sourceRef` values may carry `${input.x}` tokens that the bundle-apply path resolves against the session's `customConfig.manifestInputs`.

BindModelStrategyDto

Property	Type	Required	Description
`strategyId`	string	yes	Strategy ID to bind to the project

CatalogModelDto

Property	Type	Required	Description
`capabilities`	array of string	yes	Model capabilities
`contextWindow`	number	yes	Context window in tokens
`costInput`	number	no	Cost per 1M input tokens (USD)
`costOutput`	number	no	Cost per 1M output tokens (USD)
`deprecated`	boolean	yes	Whether the model is deprecated
`displayName`	string	yes	Display name
`family`	string	yes	Model family
`modelId`	string	yes	Model identifier
`outputWindow`	number	no	Output window in tokens
`popular`	boolean	yes	Whether this is a popular model
`reasoning`	boolean	yes	Whether the model supports reasoning
`releaseDate`	string	no	Release date (ISO)
`vision`	boolean	yes	Whether the model supports vision

CatalogModelDtoDataArrayEnvelope

Property	Type	Required	Description
`data`	array of CatalogModelDto	yes

CatalogProviderDto

Property	Type	Required	Description
`apiType`	one of `"openai-compatible"`, `"anthropic"`, `"google"`	yes	SDK / API type
`authType`	one of `"bearer"`, `"api-key-header"`	yes
`baseUrl`	string	yes	Default base URL
`description`	string	yes
`keyHelpUrl`	string	yes
`keyPlaceholder`	string	yes
`logoUrl`	string	yes	Logo URL (SVG)
`name`	string	yes	Display name
`slug`	string	yes	Provider slug
`syncSupported`	boolean	yes	Whether live /models sync is supported
`tier`	one of `"featured"`, `"popular"`	yes

CatalogProviderDtoDataArrayEnvelope

Property	Type	Required	Description
`data`	array of CatalogProviderDto	yes

CodeStandardResponseDto

Property	Type	Required	Description
`category`	string	yes	Standard category
`content`	string	yes	Standard content
`createdAt`	string (date-time)	yes
`createdBy`	string	yes	Creator ID
`id`	string	yes	Unique ID
`isActive`	boolean	yes	Whether the standard is active
`name`	string	yes	Standard name
`orgId`	string	yes	Organization ID
`projectId`	string	no	Project ID (null for org-wide standards)
`updatedAt`	string (date-time)	yes

CodeStandardResponseDtoDataArrayEnvelope

Property	Type	Required	Description
`data`	array of CodeStandardResponseDto	yes

CodeStandardResponseDtoDataEnvelope

Property	Type	Required	Description
`data`	CodeStandardResponseDto	yes

CreateAgentDto

Property	Type	Required	Description
`abilities`	array of string	no	Special-ability CapabilityRefs granted on top of the reach tier; unioned into the auto-grant manifest.
`capability`	AgentCapabilityLayerDto	no	Capability envelope applied at runtime. When omitted, a read-only default is applied at publish time.
`description`	string	no
`displayName`	string	yes
`durableWorkspace`	DurableWorkspaceRefDto or null	no	Declares that every session this agent launches attaches to ONE durable, NAMED workspace shared across independent sessions/pods (multi-pod `WorkspaceSharingPolicy.Shared`) keyed by `workspaceRef` — its files are the source of truth and cross-session mutating turns serialize through a write-lease. Omitted / `null` ⇒ a private per-session workspace (the default).
`internetEgressCeiling`	WorkerInternetEgress	no	Ceiling on whether the worker pod running this agent may open sockets to the public internet. A ceiling the agent ACCEPTS, never a grant: the effective mode is the most restrictive of the deployment, the owning Space, this value and the session, so `open` here widens nothing. Omitted ⇒ the agent states nothing and no default is substituted.
`reachTier`	AgentReachTier	no	The agent's permission ceiling. Omitted ⇒ `as-owner` (the agent's authority equals its owner's — today's acts-as behavior). Enforcement lives in authz; this service only owns + exposes the field.
`root`	AgentNodeDto	yes	The recursive root (primary agent) node.
`scope`	ScopeSource	yes	Owner tier. Customer-authorable tiers only — `org` (default) or `user`. `project` is reached via re-scope; `biome`/`system` are platform-seeded.
`sessionLifecycle`	AgentSessionLifecycle	no	The agent's default session lifecycle — `ephemeral` (idle-auto-pauses) or `persistent` (always-on). A launched session inherits this unless the launch passes an explicit override. Omitted ⇒ the agent declares no preference and agent-session-api applies its own `ephemeral` default.
`slug`	string	yes	Stable slug; lowercase, hyphenated.
`standingSession`	StandingSessionRefDto or null	no	Declares that every launch of this agent runs in ONE reusable "standing" session keyed by `pinKey` — agent-session-api adopts a non-terminal existing session for `(orgId, projectId, pinKey)` instead of minting a fresh one. Omitted / `null` ⇒ a fresh session per launch (the default).
`workspace`	AgentWorkspaceDto	no	Authored workspace-layout config — persist-paths + per-slot overrides, env vars, output surface. Optional + additive: a composition with no workspace block runs on the AWP base slots alone. Projected onto the run's WorkspaceSpec by the resolution service.
`workspaceSharing`	WorkspaceSharingPolicy	no	Whether a session launch of this agent may POOL onto a compatible, already-running pod as a new thread (`shareable`) or must always run in its own isolated worker (`isolated`). Omitted ⇒ the DB default `isolated` (the fail-safe) — mark `shareable` ONLY for workspace-stateless / chat-style agents that never mutate `/workspace`.

CreateAgentKernelDto

Property	Type	Required	Description
`avatarSeed`	string	no	DiceBear avatar seed (defaults to slug).
`avatarStyle`	string	no	DiceBear avatar style family.
`description`	string	no	Free-text description rendered on the agent profile.
`displayName`	string	yes	Display name for the agent.
`forkedFrom`	string	no	Origin reference set when this row was forked from a biome/system agent. Format: `<source>:<biomeId\|"">:<slug>`. Survives uninstall — UI surfaces "fork without origin" if the origin is gone.
`mcpServices`	array of string	no	MCP service keys this agent can access.
`modelLane`	string	yes	The model LANE this agent routes on — rules select on it. An OPEN vocabulary: every key is declared by a `model-lane` contribution, and `GET /work-kinds?kind=model-lane` enumerates what this organisation may use. Distinct from AgentRunRole (permissions).
`scope`	ScopeSource	no	Owner tier the agent is created under. Customer-authorable tiers only — `org` (the org owns it; default) or `user` (private to the creator, promotable later via re-scope). `biome` and `system` rows are seeded by the platform, never created through this endpoint; `project` rows are created via re-scope from `user`/`org`.
`skills`	array of string	no	Open-world skill tags.
`slug`	string	yes	Unique slug for the agent definition.
`systemPrompt`	string	yes	Base system prompt.
`temperature`	number	no	LLM temperature override.

CreateCodeStandardDto

Property	Type	Required	Description
`category`	one of `"code_style"`, `"api_conventions"`, `"testing"`, `"security"`, `"naming"`, `"documentation"`	yes
`content`	string	yes
`isActive`	boolean	no
`name`	string	yes
`projectId`	string or null	no	Project ID — null for org-wide standard

CreateModelDto

Property	Type	Required	Description
`capabilities`	array of string	no	Model capabilities
`contextWindow`	number	yes	Required. Max input tokens the model supports (no silent fallback).
`displayName`	string	yes
`embeddingDimension`	number	no	Embedding vector dimension (for embedding models only)
`isActive`	boolean	no
`modalities`	array of ModelModality	no	Closed-enum INPUT modalities the model accepts, recorded with source `operator`. Omitted or empty means unknown (a provider sync may fill it). Consumed by the model-modality capability provider-binding.
`modelId`	string	yes

CreateModelResolutionRuleDto

Property	Type	Required	Description
`priority`	number	no	Tie-break hint when two rules match the same number of dimensions and the fixed dimension-priority ladder does not separate them. Higher wins. Defaults to 0.
`selector`	ModelResolutionSelectorDto	no	Dimensional selector. At least one dimension is required: an empty selector is refused (set the org default via the org model default).
`targetModelId`	string	yes	Registry model record id (LlmModel.id) the rule routes to. It must be an active model on an active provider the organisation or the platform owns; otherwise the write is refused with 422 MODEL_NOT_SERVED.
`targetReasoningEffort`	string	no	Optional reasoning effort carried by the resolved model, forwarded verbatim to the provider. OPEN vocabulary owned by the provider (e.g. minimal/low/medium/high, or low/medium/high/xhigh/max) — not an enum here, so a new provider spelling is not refused at the door.
`targetTemperature`	number	no	Optional temperature override carried by the resolved model.

CreateModelStrategyDto

Property	Type	Required	Description
`description`	string	no
`displayName`	string	yes
`rules`	array of CreateModelResolutionRuleDto	yes	The strategy's rules. A strategy holds only rules: each has the same shape and validation as an org-level model resolution rule, and applies to the projects this strategy is bound to.
`slug`	string	yes

CreateProjectAgentOverrideDto

Property	Type	Required	Description
`agentIdentityId`	string	yes	Stable AgentIdentity ID to override for this project.
`mcpServicesAdd`	array of string	no	Additional MCP services to enable.
`mcpServicesRemove`	array of string	no	MCP services to disable for this project.
`systemPromptAppend`	string	no	Text appended to the base system prompt.
`temperature`	number	no

CreateProviderDto

Property	Type	Required	Description
`apiKey`	string	no	Optional enrolment transport: when supplied, the key is sealed in the credential plane and becomes the organisation's default credential (org_default pointer) for this provider's key; the provider row never stores it. Omit it when members use their own keys.
`apiType`	one of `"openai-compatible"`, `"anthropic"`, `"google"`	yes
`authType`	one of `"bearer"`, `"api-key-header"`, `"none"`	no	How the provider authenticates upstream. Omitted = `bearer`. `none` declares a keyless endpoint (openai-compatible only): `apiKey` may then not be supplied, and no credential header is sent.
`baseUrl`	string	yes
`catalogSlug`	string	no	Catalog provider slug — used to seed known models on creation
`maxConcurrentRequests`	number	no	Maximum concurrent requests allowed to this provider. Null = unlimited.
`name`	string	yes
`slug`	string	yes	URL-safe identifier

CredentialKind

Type: one of "llm-provider", "scm-token", "oauth-bearer", "mcp-auth", "wallet-entry", "generic-secret"

DelegatedAgentRefDto

Property	Type	Required	Description
`slug`	string	yes
`version`	string	yes	Generated revision number (decimal string).

DelegatedAgentSubjectDto

Property	Type	Required	Description
`id`	string	yes
`kind`	one of `"agent"`	yes

DelegatedApertureConstraintsDto

Property	Type	Required	Description
`apertures`	array of AgentCapabilityApertureDto	yes
`boundaries`	array of AgentCapabilitySetExpressionDto	yes

DurableWorkspaceRefDto

Property	Type	Required	Description
`workspaceRef`	string	yes	Stable identity of the durable shared workspace. Every session of every agent that resolves this ref within the same `(orgId, projectId)` mounts the SAME durable `/workspace` (its files are the source of truth; a session is a disposable cache).

EmbeddingCommitmentActivationDto

Property	Type	Required	Description
`activated`	boolean	yes	True when this call flipped the commitment proposed → active; false when it was already active.

EmbeddingCommitmentActivationDtoDataEnvelope

Property	Type	Required	Description
`data`	EmbeddingCommitmentActivationDto	yes

EmbeddingCommitmentDto

Property	Type	Required
`committedAt`	string (date-time) or null	no
`createdAt`	string (date-time)	yes
`createdBy`	string	yes
`dimension`	number	yes
`id`	string	yes
`modelId`	string	yes
`orgId`	string	yes
`pendingMigration`	EmbeddingMigrationDto or null	no
`provider`	EmbeddingProviderSummaryDto	yes
`providerId`	string	yes
`state`	one of `"proposed"`, `"active"`, `"migrating"`	yes
`updatedAt`	string (date-time)	yes

EmbeddingCommitmentDtoDataEnvelope

Property	Type	Required	Description
`data`	EmbeddingCommitmentDto	yes

EmbeddingConsumer

Type: one of "search_api", "memory_api"

EmbeddingDimensionVerdict

What this deployment's vector storage would do with this model's dimension. indexed is fully served; stored_unindexed works and is a linear scan; unsupported cannot be stored and must not be offered; unverified means NO consumer has declared what it can hold — which is an ANSWER, not a synonym for "fine". A client MUST render this rather than deciding from a local list of dimensions: a local list is how 4096 came to be offered with no storage behind it.

Type: one of "indexed", "stored_unindexed", "unsupported", "unverified"

EmbeddingMigrationDto

Property	Type	Required
`approvedAt`	string (date-time)	yes
`approvedBy`	string	yes
`completedAt`	string (date-time) or null	no
`consumerStates`	array of MigrationConsumerProgressDto	yes
`failedAt`	string (date-time) or null	no
`failureReason`	string or null	no
`fromDimension`	number	yes
`fromModelId`	string	yes
`fromProviderId`	string	yes
`id`	string	yes
`orgId`	string	yes
`rolledBackAt`	string (date-time) or null	no
`startedAt`	string (date-time) or null	no
`state`	one of `"approved"`, `"in_progress"`, `"completed"`, `"failed"`, `"rolled_back"`	yes
`toDimension`	number	yes
`toModelId`	string	yes
`toProviderId`	string	yes

EmbeddingMigrationDtoDataEnvelope

Property	Type	Required	Description
`data`	EmbeddingMigrationDto	yes

EmbeddingMigrationRollbackDto

Property	Type	Required	Description
`rolledBack`	boolean	yes	Always true — the rollback either completed or the request failed with an error envelope.

EmbeddingMigrationRollbackDtoDataEnvelope

Property	Type	Required	Description
`data`	EmbeddingMigrationRollbackDto	yes

EmbeddingModelOptionDto

Property	Type	Required	Description
`capabilities`	array of string	yes
`contextWindow`	number	yes
`dimensionVerdict`	EmbeddingDimensionVerdict	yes	What this deployment's vector storage would do with this model's dimension. `indexed` is fully served; `stored_unindexed` works and is a linear scan; `unsupported` cannot be stored and must not be offered; `unverified` means NO consumer has declared what it can hold — which is an ANSWER, not a synonym for "fine". A client MUST render this rather than deciding from a local list of dimensions: a local list is how 4096 came to be offered with no storage behind it.
`displayName`	string	yes
`embeddingDimension`	number	yes
`id`	string	yes
`modelId`	string	yes
`provider`	EmbeddingProviderSummaryDto	yes

EmbeddingModelOptionDtoDataArrayEnvelope

Property	Type	Required	Description
`data`	array of EmbeddingModelOptionDto	yes

EmbeddingProviderSummaryDto

Property	Type	Required
`apiType`	string	yes
`baseUrl`	string	yes
`id`	string	yes
`isSystem`	boolean	yes
`name`	string	yes
`slug`	string	yes

EmbeddingStorageDeclarationAckDto

Property	Type	Required	Description
`declaredAt`	string (date-time)	yes	When THIS service recorded it, from its own clock rather than the caller's. It is what makes a consumer that stopped re-declaring visible: its row simply stops moving.
`recorded`	boolean	yes

EmbeddingStorageDeclarationAckDtoDataEnvelope

Property	Type	Required	Description
`data`	EmbeddingStorageDeclarationAckDto	yes

EmbeddingStorageDeclarationDto

Property	Type	Required	Description
`backend`	string	yes
`backendVersion`	string	yes	Provenance for the sets above. A bare capability is a fact with no source; naming the backend and version makes it checkable, and makes a backend change visible instead of silently keeping an old answer.
`consumer`	EmbeddingConsumer	yes
`indexedDimensions`	array of number	yes	The SUBSET of `storableDimensions` this consumer can also serve from an ANN index. Outside it recall is an exact scan: correct, and slower. It is advisory and never causes a refusal — conflating the two is what made 4096 look unsupported when it was merely unindexable. May be empty, which honestly means nothing is ANN-indexed.
`storableDimensions`	array of number	yes	EXACT SET of dimensions this consumer can store — every one backed by real storage it has verified, not an interval. A dimension absent from this list is `unsupported`, including one that falls between two members. Must be non-empty: a consumer that cannot establish its own capability declares NOTHING and is reported `unverified`, which is a different answer from "stores nothing" and refuses nothing.

ExactAgentRevisionAuthorityResponseDto

Property	Type	Required	Description
`capabilityAperture`	AgentCapabilityApertureDto	yes
`contentHash`	string (sha256)	yes
`delegatedApertures`	array of ResolvedDelegatedApertureDto	yes	Every agent reachable by DELEGATION from the primary agent (composition children and intrinsic delegates, cycles resolved to a fixed point, bounded by the platform delegation depth), each with its effective authority `parent ∩ own ∩ boundary`. Sorted by `agentKey`; exactly one entry (the primary agent) has a null parent.
`internetEgressCeiling`	WorkerInternetEgress	no	Ceiling on whether the worker pod running the PRIMARY agent may open sockets to the public internet. A ceiling the agent ACCEPTS, never a grant: the effective mode is the most restrictive of the deployment, the owning Space, this value and the session, so `open` here widens nothing. RESOLVED: the agent's own declaration joined along its `extends` chain, so a child never widens its base. ABSENT when nothing in the chain declares one — there is no default.
`ownerSpaceRef`	AgentOwnerSpaceRefDto	yes
`preparationGenerationId`	string	yes
`reachTier`	one of `"sandbox"`, `"as-owner"`, `"connected"`, `"full"`	yes
`resolvedGraph`	ExactAgentRevisionResolvedGraphDto	yes
`resourceId`	string	yes
`revisionId`	string	yes

ExactAgentRevisionAuthorityResponseDtoDataEnvelope

Property	Type	Required	Description
`data`	ExactAgentRevisionAuthorityResponseDto	yes

ExactAgentRevisionResolvedGraphDto

Property	Type	Required	Description
`root`	AgentNodeDto	yes

InputArtifactEntryDto

Property	Type	Required
`contentType`	string	no
`path`	string	yes
`sourceArtifactId`	string	yes
`sourceVersion`	number	yes

ListedModelProviderDto

Property	Type	Required	Description
`authType`	one of `"bearer"`, `"api-key-header"`, `"none"`	yes	How the provider authenticates upstream. `none` = a keyless endpoint.
`id`	string	yes
`isActive`	boolean	yes	Whether the provider is active. An inactive provider serves no model.
`name`	string	yes
`slug`	string	yes

ListedModelResponseDto

Property	Type	Required	Description
`audioInputPerSecond`	number or null	no	Per-input-audio-second price (USD per second).
`cacheReadPerMToken`	number or null	no	Cache-read price (USD per 1M tokens).
`cacheWritePerMToken`	number or null	no	Cache-write price (USD per 1M tokens).
`capabilities`	array of string	yes
`contextWindow`	object or null	no
`contextWindowSource`	ModelFactSource or null	yes	Where a model size fact came from, highest rank first: `operator` (a person set it; no provider sync ever overwrites it), `provider` (provider discovery reported it), `curated_catalog` (the curated per-provider catalogue). For `contextWindow`, null = provenance not recorded (a model registered before sources were tracked).
`contextWindowVerifiedAt`	string (date-time) or null	yes	When the source of `contextWindow` last asserted this value (ISO 8601). Null exactly when its source is null.
`createdAt`	string	yes
`deprecated`	boolean	yes	Model has rolled out of the current window.
`deprecationDate`	string or null	no	The operator-set retirement instant (ISO 8601); null when the model is not retired. From it on, resolutions that pick this model serve `successorModelId` instead, or refuse with `MODEL_RETIRED_NO_SUCCESSOR` when there is none. No catalogue sync writes it.
`displayName`	string	yes
`embeddingDimension`	number or null	no	Embedding output vector dimension (only for embedding models).
`factConflicts`	array of ModelFactConflictDto	yes	Every violated size-fact bound, with both sides' sources. Empty when consistent.
`factsConsistent`	boolean	yes	False when the effective size facts conflict (see `factConflicts`). An inconsistent model is refused for selection and use with `MODEL_FACTS_CONFLICT` until an admin changes or clears the overriding value.
`id`	string	yes
`imageInputPerUnit`	number or null	no	Per-input-image price (USD per unit).
`inputPricePerMToken`	number or null	no	Input price (USD per 1M tokens).
`isActive`	boolean	yes
`maxInputTokens`	number or null	yes	Max prompt (input) tokens, when the service caps input below the context window. Null when no source declares a cap (only the context window applies).
`maxInputTokensSource`	ModelFactSource or null	yes	Where a model size fact came from, highest rank first: `operator` (a person set it; no provider sync ever overwrites it), `provider` (provider discovery reported it), `curated_catalog` (the curated per-provider catalogue). For `maxInputTokens`, null = the fact is unknown.
`maxInputTokensVerifiedAt`	string (date-time) or null	yes	When the source of `maxInputTokens` last asserted this value (ISO 8601). Null exactly when its source is null.
`maxOutputTokens`	number or null	no	Max generation/output tokens. Bounded by contextWindow, which is the TOTAL budget (prompt + reasoning + answer). Null when no authoritative source declares one.
`maxOutputTokensSource`	ModelFactSource or null	yes	Where a model size fact came from, highest rank first: `operator` (a person set it; no provider sync ever overwrites it), `provider` (provider discovery reported it), `curated_catalog` (the curated per-provider catalogue). For `maxOutputTokens`, null with a null value = the fact is unknown; null beside a value = provenance not recorded (a model registered before sources were tracked).
`maxOutputTokensVerifiedAt`	string (date-time) or null	yes	When the source of `maxOutputTokens` last asserted this value (ISO 8601). Null exactly when its source is null.
`modalities`	array of ModelModality	yes	Closed-enum INPUT modalities the model accepts — a capability fact (see `modalitiesSource`). Empty when nobody declared any.
`modalitiesSource`	ModelFactSource or null	yes	Where the `modalities` capability fact came from, highest rank first: `operator` (no sync overwrites it), `provider` (the provider declared it), `curated_catalog`. an empty set with a null source = the fact is unknown (nobody declared it); a non-default value with a null source = provenance not recorded.
`modalitiesVerifiedAt`	string (date-time) or null	yes	When the source of `modalities` last asserted it (ISO 8601). Null exactly when its source is null.
`mode`	ModelMode or null	no	Operating mode of the model. Null when no signal resolved one.
`modelId`	string	yes
`outputPricePerMToken`	number or null	no	Output price (USD per 1M tokens).
`priceCurrency`	string or null	no	Currency of the price columns ('USD' when any price is present, null otherwise).
`provider`	ListedModelProviderDto	yes
`providerId`	string	yes
`servable`	boolean	yes	True iff model resolution would serve this model to the organisation: the model is active, its provider is active and usable by the org (the same predicate resolution refuses with `MODEL_NOT_SERVED`), and its size facts agree (else resolution refuses with `MODEL_FACTS_CONFLICT`). Model pickers offer only servable models.
`successorModelId`	string or null	no	The model record id resolutions serve instead of this one from `deprecationDate`; null when none.
`supportsAudioInput`	boolean	yes	Model accepts audio inputs.
`supportsAudioInputSource`	ModelFactSource or null	yes	Where the `supportsAudioInput` capability fact came from, highest rank first: `operator` (no sync overwrites it), `provider` (the provider declared it), `curated_catalog`. `false` with a null source = the fact is unknown (nobody declared it); a non-default value with a null source = provenance not recorded.
`supportsAudioInputVerifiedAt`	string (date-time) or null	yes	When the source of `supportsAudioInput` last asserted it (ISO 8601). Null exactly when its source is null.
`supportsJsonMode`	boolean	yes	Model supports structured JSON output mode.
`supportsJsonModeSource`	ModelFactSource or null	yes	Where the `supportsJsonMode` capability fact came from, highest rank first: `operator` (no sync overwrites it), `provider` (the provider declared it), `curated_catalog`. `false` with a null source = the fact is unknown (nobody declared it); a non-default value with a null source = provenance not recorded.
`supportsJsonModeVerifiedAt`	string (date-time) or null	yes	When the source of `supportsJsonMode` last asserted it (ISO 8601). Null exactly when its source is null.
`supportsPdfInput`	boolean	yes	Model accepts PDF inputs.
`supportsPdfInputSource`	ModelFactSource or null	yes	Where the `supportsPdfInput` capability fact came from, highest rank first: `operator` (no sync overwrites it), `provider` (the provider declared it), `curated_catalog`. `false` with a null source = the fact is unknown (nobody declared it); a non-default value with a null source = provenance not recorded.
`supportsPdfInputVerifiedAt`	string (date-time) or null	yes	When the source of `supportsPdfInput` last asserted it (ISO 8601). Null exactly when its source is null.
`supportsPromptCaching`	boolean	yes	Model supports prompt caching.
`supportsPromptCachingSource`	ModelFactSource or null	yes	Where the `supportsPromptCaching` capability fact came from, highest rank first: `operator` (no sync overwrites it), `provider` (the provider declared it), `curated_catalog`. `false` with a null source = the fact is unknown (nobody declared it); a non-default value with a null source = provenance not recorded.
`supportsPromptCachingVerifiedAt`	string (date-time) or null	yes	When the source of `supportsPromptCaching` last asserted it (ISO 8601). Null exactly when its source is null.
`supportsReasoning`	boolean	yes	Model performs explicit reasoning.
`supportsReasoningSource`	ModelFactSource or null	yes	Where the `supportsReasoning` capability fact came from, highest rank first: `operator` (no sync overwrites it), `provider` (the provider declared it), `curated_catalog`. `false` with a null source = the fact is unknown (nobody declared it); a non-default value with a null source = provenance not recorded.
`supportsReasoningVerifiedAt`	string (date-time) or null	yes	When the source of `supportsReasoning` last asserted it (ISO 8601). Null exactly when its source is null.
`supportsTools`	boolean	yes	Model exposes function/tool calling.
`supportsToolsSource`	ModelFactSource or null	yes	Where the `supportsTools` capability fact came from, highest rank first: `operator` (no sync overwrites it), `provider` (the provider declared it), `curated_catalog`. `false` with a null source = the fact is unknown (nobody declared it); a non-default value with a null source = provenance not recorded.
`supportsToolsVerifiedAt`	string (date-time) or null	yes	When the source of `supportsTools` last asserted it (ISO 8601). Null exactly when its source is null.
`supportsVision`	boolean	yes	Model accepts image inputs.
`supportsVisionSource`	ModelFactSource or null	yes	Where the `supportsVision` capability fact came from, highest rank first: `operator` (no sync overwrites it), `provider` (the provider declared it), `curated_catalog`. `false` with a null source = the fact is unknown (nobody declared it); a non-default value with a null source = provenance not recorded.
`supportsVisionVerifiedAt`	string (date-time) or null	yes	When the source of `supportsVision` last asserted it (ISO 8601). Null exactly when its source is null.
`supportsWebSearch`	boolean	yes	Model supports built-in web search.
`supportsWebSearchSource`	ModelFactSource or null	yes	Where the `supportsWebSearch` capability fact came from, highest rank first: `operator` (no sync overwrites it), `provider` (the provider declared it), `curated_catalog`. `false` with a null source = the fact is unknown (nobody declared it); a non-default value with a null source = provenance not recorded.
`supportsWebSearchVerifiedAt`	string (date-time) or null	yes	When the source of `supportsWebSearch` last asserted it (ISO 8601). Null exactly when its source is null.
`tieredPricing`	array of ModelPricingTierDto or null	no	Context-tier pricing overlay sorted ascending by minInputTokens. Null when no tier is published.

ListedModelResponseDtoDataArrayEnvelope

Property	Type	Required	Description
`data`	array of ListedModelResponseDto	yes

LlmReadinessDetailsDto

Property	Type	Required	Description
`missingModelSlugs`	array of string	no	Model slugs that need to be synced
`missingProviderSlugs`	array of string	no	Provider slugs that need to be added
`strategyId`	string	no	Related strategy ID
`strategyName`	string	no	Related strategy name
`unresolvedLanes`	array of UnresolvedLaneDto	no	Present on `required_lane_unresolved`: every agent whose own model lane resolves to no usable model at the scope, ordered by agent slug then lane. The agents considered are those from a biome enabled for the organisation, not hidden, and supporting agent sessions. Agent-lane resolution only: a launch definition's own model intent and project scope are judged at launch, not here.

LlmReadinessResponseDto

Property	Type	Required	Description
`action`	one of `"add_provider"`, `"sync_models"`, `"set_default_model"`, `"fix_routed_models"`, `"fix_project_strategy"`, `"bind_required_lanes"` or null	no	Suggested remediation action. `bind_required_lanes` means an agent's own model lane has no resolution at the requested scope; it does not speak for a launch definition's own model intent, which is judged at launch.
`details`	LlmReadinessDetailsDto	no	Additional diagnostic details
`message`	string	yes	Human-readable status message
`ready`	boolean	yes	Whether LLM is ready to use
`status`	one of `"ready"`, `"missing_provider"`, `"missing_models"`, `"missing_model_routing"`, `"routed_models_unavailable"`, `"project_strategy_unavailable"`, `"required_lane_unresolved"`	yes	`required_lane_unresolved` reports agent-lane resolution at the requested scope only (each agent's own model lane); a launch definition's own model intent and project scope are judged at launch, not here.

LlmReadinessResponseDtoDataEnvelope

Property	Type	Required	Description
`data`	LlmReadinessResponseDto	yes

MigrationAcknowledgementDto

Property	Type	Required	Description
`acknowledged`	boolean	yes	Always true — the report was recorded. A failure surfaces as an error envelope, never as `false`.

MigrationAcknowledgementDtoDataEnvelope

Property	Type	Required	Description
`data`	MigrationAcknowledgementDto	yes

MigrationConsumerDelegationDto

Property	Type	Required	Description
`carrierId`	string	yes	The carrier bound to the child delegation: what the consumer mints each delegated context from, and renews on its own tick.
`consumer`	one of `"search_api"`, `"memory_api"`	yes
`delegationId`	string	yes	The consumer’s child delegation id. Lends model:gateway-session.mint@1 for the org, presentable only by this consumer’s workload.
`state`	MigrationConsumerState	yes	This consumer’s own progress on the migration. A consumer whose state is `completed` has finished its part and must not provision it again: provisioning a completed consumer is refused with a 409.

MigrationConsumerDoneDto

Property	Type	Required	Description
`consumer`	one of `"search_api"`, `"memory_api"`	yes	Which consumer has finished its backfill.

MigrationConsumerFailedDto

Property	Type	Required	Description
`consumer`	one of `"search_api"`, `"memory_api"`	yes
`reason`	string	yes	Reason the consumer gave up.

MigrationConsumerProgressDto

Property	Type	Required
`consumer`	one of `"search_api"`, `"memory_api"`	yes
`failedItems`	number	yes
`failureReason`	string or null	no
`id`	string	yes
`lastHeartbeatAt`	string (date-time) or null	no
`lastProcessedId`	string or null	no
`processedItems`	number	yes
`sourceCursor`	string or null	no
`state`	one of `"pending"`, `"provisioning"`, `"dual_writing"`, `"completed"`, `"failed"`	yes
`totalItems`	number	yes

MigrationConsumerState

This consumer’s own progress on the migration. A consumer whose state is completed has finished its part and must not provision it again: provisioning a completed consumer is refused with a 409.

Type: one of "pending", "provisioning", "dual_writing", "completed", "failed"

MigrationHeartbeatDto

Property	Type	Required	Description
`consumer`	one of `"search_api"`, `"memory_api"`	yes	Which consumer is reporting progress.
`failedDelta`	number	no	Number of items that failed since the last heartbeat.
`lastProcessedId`	string	yes	Last source row id processed.
`processedDelta`	number	yes	Number of new items processed since the last heartbeat.

MigrationProvisionDto

Property	Type	Required	Description
`consumer`	one of `"search_api"`, `"memory_api"`	yes	Which consumer is provisioning.
`sourceCursor`	string	yes	High-water-mark cursor snapshot taken at provision time.
`totalItems`	number	yes	Total source items the consumer plans to process.

MigrationTargetDto

Property	Type	Required	Description
`consumers`	array of MigrationConsumerDelegationDto	yes	The durable CHILD delegation each consumer re-embeds under, lent by the migration’s root delegation (anchored on the migration, originated by the approving admin at migration start). A consumer mints a delegated context per use from ITS OWN entry (a delegation names exactly one presenting workload) with mintDurableDelegatedToken(identity, { delegationId, organizationId, carrierId }), never caches it, and renews that carrier on its own tick (DurableDelegationHolderClient.renewCarrier). Entries with no child yet are omitted. Every entry ends when the migration completes, fails or is rolled back: authorization-api ends the delegations of an anchor that is no longer active.
`migrationId`	string	yes
`toDimension`	number	yes
`toModelId`	string	yes
`toProviderApiType`	string	yes
`toProviderBaseUrl`	string	yes
`toProviderId`	string	yes
`toProviderIsSystem`	boolean	yes
`toProviderSlug`	string	yes

ModelCapability

A modality this invocation REQUIRES — the Matrix modelCapability dimension, so one rule can route every vision-needing invocation to a multimodal model without pinning a provider. It is ALSO a requirement on whatever the resolution selects: a resolved model that does not provide it is refused with 422 MODEL_CAPABILITY_UNSUPPORTED rather than returned, and no other model is substituted. Closed kernel set; an unknown value is a 400.

Type: one of "vision", "long_context", "reasoning", "function_calling", "audio", "embedding"

ModelFactConflictDto

Property	Type	Required	Description
`fact`	ModelFactName	yes	A model size fact (`ModelResponseDto` property name). Here: the capped fact (`maxInputTokens` or `maxOutputTokens`) that exceeds its bound.
`limitFact`	ModelFactName	yes	The bounding fact (always `contextWindow` today).
`limitSource`	ModelFactSource or null	yes	Where the bound came from; null = provenance not recorded.
`limitValue`	number	yes	Its effective value.
`source`	ModelFactSource or null	yes	Where that value came from; null = provenance not recorded.
`value`	number	yes	Its effective value.

ModelFactName

A model size fact (ModelResponseDto property name). Here: the capped fact (maxInputTokens or maxOutputTokens) that exceeds its bound.

Type: one of "contextWindow", "maxInputTokens", "maxOutputTokens"

ModelFactSource

Where a model size fact came from, highest rank first: operator (a person set it; no provider sync ever overwrites it), provider (provider discovery reported it), curated_catalog (the curated per-provider catalogue). For contextWindow, null = provenance not recorded (a model registered before sources were tracked).

Type: one of "operator", "provider", "curated_catalog"

ModelModality

Closed-enum INPUT modalities the model accepts — a capability fact (see modalitiesSource). Empty when nobody declared any.

Type: one of "text", "image", "pdf", "audio", "video"

ModelMode

Operating mode of the model. Null when no signal resolved one.

Type: one of "chat", "embedding"

ModelPricingTierDto

Property	Type	Required	Description
`cacheReadPerMToken`	number	no	Cache-read price (USD per 1M tokens), when the band declares one.
`cacheWritePerMToken`	number	no	Cache-write price (USD per 1M tokens), when the band declares one.
`inputPricePerMToken`	number	yes	Input price (USD per 1M tokens) within this band.
`minInputTokens`	number	yes	Lower input-token bound this band applies from.
`outputPricePerMToken`	number	yes	Output price (USD per 1M tokens) within this band.

ModelResolutionErrorCode

The code a launch of this agent is refused with — the same value the resolve doors put on the refusal.

Type: one of "MODEL_NOT_CONFIGURED", "MODEL_NOT_SERVED", "MODEL_TARGET_NOT_A_MODEL", "MODEL_PROVIDER_PIN_MISMATCH", "MODEL_FACTS_CONFLICT", "MODEL_CAPABILITY_UNSUPPORTED", "MODEL_RETIRED_NO_SUCCESSOR"

ModelResolutionPreviewOutcome

Type: one of "resolved", "refused"

ModelResolutionPreviewResponseDto

Property	Type	Required	Description
`outcome`	ModelResolutionPreviewOutcome	yes
`refusal`	ModelResolutionRefusalDto	no	Why resolution refused. Present exactly when `outcome` is `refused`.
`resolved`	ResolvedPhaseConfigDto	no	The resolved model. Present exactly when `outcome` is `resolved`.

ModelResolutionPreviewResponseDtoDataEnvelope

Property	Type	Required	Description
`data`	ModelResolutionPreviewResponseDto	yes

ModelResolutionRefusalCode

Type: one of "MODEL_NOT_CONFIGURED", "MODEL_NOT_SERVED", "MODEL_RETIRED_NO_SUCCESSOR"

ModelResolutionRefusalDto

Property	Type	Required	Description
`code`	ModelResolutionRefusalCode	yes
`message`	string	yes	What is missing and how to fix it.

ModelResolutionRuleResponseDto

Property	Type	Required	Description
`createdAt`	string	yes
`createdBy`	string	yes
`id`	string	yes
`managedBy`	string or null	no	ResourceManagedBy ownership KIND (`ui`/`iac`/`seeder`/`runtime_registered`/`system`). NULL for a plain user/org-admin-authored rule.
`managedKey`	string or null	no	The owner's stable natural key. NULL for authored rules; non-null for managed/reconciler rows.
`managedOwner`	string or null	no	Owning service identity for a service-actor-upserted rule (derived from the service subject). NULL for authored/reconciler rules.
`orgId`	string	yes
`priority`	number	yes
`selector`	ModelResolutionSelectorResponseDto	yes
`strategyId`	string or null	no	The model strategy that owns this rule; null for an org-level rule.
`targetModelId`	string	yes	Registry model record id (LlmModel.id) the rule routes to.
`targetReasoningEffort`	string or null	no
`targetTemperature`	number or null	no
`updatedAt`	string	yes

ModelResolutionRuleResponseDtoDataArrayEnvelope

Property	Type	Required	Description
`data`	array of ModelResolutionRuleResponseDto	yes

ModelResolutionRuleResponseDtoDataEnvelope

Property	Type	Required	Description
`data`	ModelResolutionRuleResponseDto	yes

ModelResolutionSelectorDto

Property	Type	Required	Description
`agent`	string	no	Match when the resolving agent slug equals this value.
`extra`	map of string	no	Forward-compat bag of additional registry dimensions → required literal. Omit / {} when only structured dimensions are used.
`modelCapability`	one of `"vision"`, `"long_context"`, `"reasoning"`, `"function_calling"`, `"audio"`, `"embedding"`	no	Match when the invocation REQUIRES this model capability/modality (the `modelCapability` dimension — one rule routes e.g. every vision-requiring invocation to a multimodal model without pinning a provider).
`modelLane`	string	no	Match when the resolving agent's work kind equals this value (the `modelLane` dimension — one rule routes e.g. every `review`-kind agent). OPEN vocabulary with no kernel core: every lane is declared by a `model-lane` contribution, so `GET /work-kinds?kind=model-lane` enumerates the lanes this organisation may use. Checked against the org's visible set on write.
`project`	string	no	Match when the invocation project id equals this value.
`purpose`	string	no	Match when the non-agent invocation purpose equals this value. Used by system flows that route LLM calls outside the agent/skill model (memory summarization, embedding, reorganization). Closed value set is owned per consumer (see `MemoryMatrixPurpose` in memory-api).
`skill`	string	no	Match when the active/launched skill slug equals this value.
`stage`	string	no	Match when the invocation pipeline stage key equals this value.

ModelResolutionSelectorResponseDto

Property	Type	Required	Description
`agent`	string or null	no
`extra`	map of string	yes	Additional registry dimensions. `{}` when none.
`modelCapability`	one of `"vision"`, `"long_context"`, `"reasoning"`, `"function_calling"`, `"audio"`, `"embedding"` or null	no
`modelLane`	string or null	no
`project`	string or null	no
`purpose`	string or null	no
`skill`	string or null	no
`stage`	string or null	no

ModelResolutionSource

Which resolution path produced this selection.

Type: one of "matrix", "override", "org_default", "unresolved"

ModelResponseDto

Property	Type	Required	Description
`audioInputPerSecond`	number or null	no	Per-input-audio-second price (USD per second).
`cacheReadPerMToken`	number or null	no	Cache-read price (USD per 1M tokens).
`cacheWritePerMToken`	number or null	no	Cache-write price (USD per 1M tokens).
`capabilities`	array of string	yes
`contextWindow`	object or null	no
`contextWindowSource`	ModelFactSource or null	yes	Where a model size fact came from, highest rank first: `operator` (a person set it; no provider sync ever overwrites it), `provider` (provider discovery reported it), `curated_catalog` (the curated per-provider catalogue). For `contextWindow`, null = provenance not recorded (a model registered before sources were tracked).
`contextWindowVerifiedAt`	string (date-time) or null	yes	When the source of `contextWindow` last asserted this value (ISO 8601). Null exactly when its source is null.
`createdAt`	string	yes
`deprecated`	boolean	yes	Model has rolled out of the current window.
`deprecationDate`	string or null	no	The operator-set retirement instant (ISO 8601); null when the model is not retired. From it on, resolutions that pick this model serve `successorModelId` instead, or refuse with `MODEL_RETIRED_NO_SUCCESSOR` when there is none. No catalogue sync writes it.
`displayName`	string	yes
`embeddingDimension`	number or null	no	Embedding output vector dimension (only for embedding models).
`factConflicts`	array of ModelFactConflictDto	yes	Every violated size-fact bound, with both sides' sources. Empty when consistent.
`factsConsistent`	boolean	yes	False when the effective size facts conflict (see `factConflicts`). An inconsistent model is refused for selection and use with `MODEL_FACTS_CONFLICT` until an admin changes or clears the overriding value.
`id`	string	yes
`imageInputPerUnit`	number or null	no	Per-input-image price (USD per unit).
`inputPricePerMToken`	number or null	no	Input price (USD per 1M tokens).
`isActive`	boolean	yes
`maxInputTokens`	number or null	yes	Max prompt (input) tokens, when the service caps input below the context window. Null when no source declares a cap (only the context window applies).
`maxInputTokensSource`	ModelFactSource or null	yes	Where a model size fact came from, highest rank first: `operator` (a person set it; no provider sync ever overwrites it), `provider` (provider discovery reported it), `curated_catalog` (the curated per-provider catalogue). For `maxInputTokens`, null = the fact is unknown.
`maxInputTokensVerifiedAt`	string (date-time) or null	yes	When the source of `maxInputTokens` last asserted this value (ISO 8601). Null exactly when its source is null.
`maxOutputTokens`	number or null	no	Max generation/output tokens. Bounded by contextWindow, which is the TOTAL budget (prompt + reasoning + answer). Null when no authoritative source declares one.
`maxOutputTokensSource`	ModelFactSource or null	yes	Where a model size fact came from, highest rank first: `operator` (a person set it; no provider sync ever overwrites it), `provider` (provider discovery reported it), `curated_catalog` (the curated per-provider catalogue). For `maxOutputTokens`, null with a null value = the fact is unknown; null beside a value = provenance not recorded (a model registered before sources were tracked).
`maxOutputTokensVerifiedAt`	string (date-time) or null	yes	When the source of `maxOutputTokens` last asserted this value (ISO 8601). Null exactly when its source is null.
`modalities`	array of ModelModality	yes	Closed-enum INPUT modalities the model accepts — a capability fact (see `modalitiesSource`). Empty when nobody declared any.
`modalitiesSource`	ModelFactSource or null	yes	Where the `modalities` capability fact came from, highest rank first: `operator` (no sync overwrites it), `provider` (the provider declared it), `curated_catalog`. an empty set with a null source = the fact is unknown (nobody declared it); a non-default value with a null source = provenance not recorded.
`modalitiesVerifiedAt`	string (date-time) or null	yes	When the source of `modalities` last asserted it (ISO 8601). Null exactly when its source is null.
`mode`	ModelMode or null	no	Operating mode of the model. Null when no signal resolved one.
`modelId`	string	yes
`outputPricePerMToken`	number or null	no	Output price (USD per 1M tokens).
`priceCurrency`	string or null	no	Currency of the price columns ('USD' when any price is present, null otherwise).
`providerId`	string	yes
`successorModelId`	string or null	no	The model record id resolutions serve instead of this one from `deprecationDate`; null when none.
`supportsAudioInput`	boolean	yes	Model accepts audio inputs.
`supportsAudioInputSource`	ModelFactSource or null	yes	Where the `supportsAudioInput` capability fact came from, highest rank first: `operator` (no sync overwrites it), `provider` (the provider declared it), `curated_catalog`. `false` with a null source = the fact is unknown (nobody declared it); a non-default value with a null source = provenance not recorded.
`supportsAudioInputVerifiedAt`	string (date-time) or null	yes	When the source of `supportsAudioInput` last asserted it (ISO 8601). Null exactly when its source is null.
`supportsJsonMode`	boolean	yes	Model supports structured JSON output mode.
`supportsJsonModeSource`	ModelFactSource or null	yes	Where the `supportsJsonMode` capability fact came from, highest rank first: `operator` (no sync overwrites it), `provider` (the provider declared it), `curated_catalog`. `false` with a null source = the fact is unknown (nobody declared it); a non-default value with a null source = provenance not recorded.
`supportsJsonModeVerifiedAt`	string (date-time) or null	yes	When the source of `supportsJsonMode` last asserted it (ISO 8601). Null exactly when its source is null.
`supportsPdfInput`	boolean	yes	Model accepts PDF inputs.
`supportsPdfInputSource`	ModelFactSource or null	yes	Where the `supportsPdfInput` capability fact came from, highest rank first: `operator` (no sync overwrites it), `provider` (the provider declared it), `curated_catalog`. `false` with a null source = the fact is unknown (nobody declared it); a non-default value with a null source = provenance not recorded.
`supportsPdfInputVerifiedAt`	string (date-time) or null	yes	When the source of `supportsPdfInput` last asserted it (ISO 8601). Null exactly when its source is null.
`supportsPromptCaching`	boolean	yes	Model supports prompt caching.
`supportsPromptCachingSource`	ModelFactSource or null	yes	Where the `supportsPromptCaching` capability fact came from, highest rank first: `operator` (no sync overwrites it), `provider` (the provider declared it), `curated_catalog`. `false` with a null source = the fact is unknown (nobody declared it); a non-default value with a null source = provenance not recorded.
`supportsPromptCachingVerifiedAt`	string (date-time) or null	yes	When the source of `supportsPromptCaching` last asserted it (ISO 8601). Null exactly when its source is null.
`supportsReasoning`	boolean	yes	Model performs explicit reasoning.
`supportsReasoningSource`	ModelFactSource or null	yes	Where the `supportsReasoning` capability fact came from, highest rank first: `operator` (no sync overwrites it), `provider` (the provider declared it), `curated_catalog`. `false` with a null source = the fact is unknown (nobody declared it); a non-default value with a null source = provenance not recorded.
`supportsReasoningVerifiedAt`	string (date-time) or null	yes	When the source of `supportsReasoning` last asserted it (ISO 8601). Null exactly when its source is null.
`supportsTools`	boolean	yes	Model exposes function/tool calling.
`supportsToolsSource`	ModelFactSource or null	yes	Where the `supportsTools` capability fact came from, highest rank first: `operator` (no sync overwrites it), `provider` (the provider declared it), `curated_catalog`. `false` with a null source = the fact is unknown (nobody declared it); a non-default value with a null source = provenance not recorded.
`supportsToolsVerifiedAt`	string (date-time) or null	yes	When the source of `supportsTools` last asserted it (ISO 8601). Null exactly when its source is null.
`supportsVision`	boolean	yes	Model accepts image inputs.
`supportsVisionSource`	ModelFactSource or null	yes	Where the `supportsVision` capability fact came from, highest rank first: `operator` (no sync overwrites it), `provider` (the provider declared it), `curated_catalog`. `false` with a null source = the fact is unknown (nobody declared it); a non-default value with a null source = provenance not recorded.
`supportsVisionVerifiedAt`	string (date-time) or null	yes	When the source of `supportsVision` last asserted it (ISO 8601). Null exactly when its source is null.
`supportsWebSearch`	boolean	yes	Model supports built-in web search.
`supportsWebSearchSource`	ModelFactSource or null	yes	Where the `supportsWebSearch` capability fact came from, highest rank first: `operator` (no sync overwrites it), `provider` (the provider declared it), `curated_catalog`. `false` with a null source = the fact is unknown (nobody declared it); a non-default value with a null source = provenance not recorded.
`supportsWebSearchVerifiedAt`	string (date-time) or null	yes	When the source of `supportsWebSearch` last asserted it (ISO 8601). Null exactly when its source is null.
`tieredPricing`	array of ModelPricingTierDto or null	no	Context-tier pricing overlay sorted ascending by minInputTokens. Null when no tier is published.

ModelResponseDtoDataArrayEnvelope

Property	Type	Required	Description
`data`	array of ModelResponseDto	yes

ModelResponseDtoDataEnvelope

Property	Type	Required	Description
`data`	ModelResponseDto	yes

ModelRetirementSubstitutionDto

Property	Type	Required	Description
`reason`	ModelSubstitutionReason	yes	Why the pick was substituted.
`retiredModel`	RetirementModelRefDto	yes	The model the rule, default or override picked; retired at resolution time.
`successorApplied`	RetirementModelRefDto	yes	The first non-retired successor, which the resolution serves.

ModelStatusDto

Property	Type	Required	Description
`deprecated`	boolean	yes	Whether the model is deprecated
`exists`	boolean	yes	Whether the model exists in the catalog
`family`	string	no	Model family
`latestInFamily`	string	no	Latest model in the family
`provider`	string	no	Provider slug

ModelStatusDtoDataEnvelope

Property	Type	Required	Description
`data`	ModelStatusDto	yes

ModelStrategyResponseDto

Property	Type	Required	Description
`createdAt`	string	yes
`createdBy`	string	yes
`description`	string or null	no
`displayName`	string	yes
`id`	string	yes
`orgId`	string	yes
`rules`	array of ModelResolutionRuleResponseDto	no	The strategy's rules: a strategy holds only rules, and they apply to the projects it is bound to.
`slug`	string	yes
`source`	ModelStrategySource	yes	Who authored this strategy. Always `org` today: the platform predefines no strategy, and no contribution kind lets a biome contribute one. A biome-contributed strategy would add a `biome` member (with its biomeId) when that contribution path exists.
`updatedAt`	string	yes
`version`	number	yes	The rule set's concurrency counter. Send it back as `expectedVersion` when replacing the whole rule set. Moves on every write that changes the rules — including a per-lane write — and not on an unrelated metadata edit.

ModelStrategyResponseDtoDataArrayEnvelope

Property	Type	Required	Description
`data`	array of ModelStrategyResponseDto	yes

ModelStrategyResponseDtoDataEnvelope

Property	Type	Required	Description
`data`	ModelStrategyResponseDto	yes

ModelStrategySource

Who authored this strategy. Always org today: the platform predefines no strategy, and no contribution kind lets a biome contribute one. A biome-contributed strategy would add a biome member (with its biomeId) when that contribution path exists.

Type: one of "org"

ModelSubstitutionReason

Why the pick was substituted.

Type: one of "model-retired"

ModelVocabularyKeyDto

Property	Type	Required	Description
`declaringBiomeIds`	array of string	yes	The biome(s) that declare this key, among the declarations visible to the caller — sorted, de-duplicated, and EVERY declarer rather than one: two biomes may declare the same key, and both are reported. Empty for a key no biome declared. PROVENANCE, NOT IDENTITY: the key alone is the stable identifier, and this list moves when a biome is renamed or replaced — group or explain by it, never key a binding on it.
`description`	string	yes	What work this key names, in the reader's words.
`key`	string	yes	The routing key. Lower-kebab and NOT namespaced by the declaring biome: a key that encoded it would move when that biome is renamed or replaced, silently invalidating every rule and strategy entry that names it.
`kind`	ModelVocabularyKind	yes	Which Matrix dimension this key may select, and which contribution kind declared it. `model-lane` — work an agent does, selecting `modelLane`. `model-purpose` — a use the platform or a biome makes of a model on nobody's behalf, selecting `purpose`.
`title`	string	yes

ModelVocabularyKind

Which Matrix dimension this key may select, and which contribution kind declared it. model-lane — work an agent does, selecting modelLane. model-purpose — a use the platform or a biome makes of a model on nobody's behalf, selecting purpose.

Type: one of "model-lane", "model-purpose"

ModelVocabularyListDto

Property	Type	Required	Description
`keys`	array of ModelVocabularyKeyDto	yes

ModelVocabularyListDtoDataEnvelope

Property	Type	Required	Description
`data`	ModelVocabularyListDto	yes

MountEntryDto

Property	Type	Required	Description
`mode`	one of `"read-only"`, `"read-write"`	yes
`mount`	string	yes
`source`	string	no	Source URI the activity resolves at mount time. Schemes: `artifact-store://`, `deliverable-spec://`, `scm://`, `literal://`. Omitted when the renderer only needs `(mount, mode)` (e.g. compose-time mount-plan summarization).

MoveRevisionPinsDto

Property	Type	Required	Description
`fromRevisionId`	string	yes	The revision pins are currently held on.
`toRevisionId`	string	yes	The revision movable pins should point at.

OrgModelDefaultResponseDto

Property	Type	Required	Description
`configured`	boolean	yes	True when the organisation has set a default model.
`modelId`	string	no	The registry model record id; present only when configured.
`setByActorId`	string	no	The actor who last set the default; present only when configured.
`updatedAt`	string	no	When the default was last set (ISO 8601); present only when configured.

OrgModelDefaultResponseDtoDataEnvelope

Property	Type	Required	Description
`data`	OrgModelDefaultResponseDto	yes

OutputSurfaceKind

Type: one of "none", "web", "static", "app", "tunnel"

PaginationMeta

Property	Type	Required
`hasMore`	boolean	no
`limit`	number	no
`nextCursor`	string or null	no
`page`	number	no
`previousCursor`	string or null	no
`total`	number	no
`totalPages`	number	no

PreviewModelResolutionDto

Property	Type	Required	Description
`modelLane`	string	yes	The model lane to resolve — the Matrix `modelLane` dimension. Must be a declared lane.
`space`	string	yes	The Space to resolve in: `xema://orgs/<id>` or `xema://orgs/<id>/projects/<id>`. It must name the organisation the caller is an admin of; it selects a Space inside it and never widens it.

ProbeProviderDto

Property	Type	Required	Description
`apiKey`	string	no	The key to probe with. Required unless `authType` is `none`.
`apiType`	one of `"openai-compatible"`, `"anthropic"`, `"google"`	yes
`authType`	one of `"bearer"`, `"api-key-header"`, `"none"`	no	Omitted = `bearer`. `none` probes a keyless endpoint (openai-compatible only): `apiKey` must then be absent and no credential header is sent.
`baseUrl`	string	yes

ProbedModelDto

Property	Type	Required	Description
`contextWindow`	number	no	TOTAL context window (prompt + answer) resolved from the provider /models response or gateway metadata — never an input cap. Absent when no authoritative total was found.
`displayName`	string	no	Human-readable model name, when the provider reports one.
`id`	string	yes	Model ID as reported by the provider /models endpoint.

ProblemActionableBy

Who can act on this refusal: user (correct the input, sign in), org-admin (an administrator of the organisation), platform (the operators of this installation), none (nobody — it cannot succeed). ABSENT when not known; never guessed.

Type: one of "user", "org-admin", "platform", "none"

ProblemDetailsDto

RFC 9457 problem document. The declared members are the reserved core; extension members are TOP-LEVEL and OPEN, owned by the refusing plane.

Property	Type	Required	Description
`actionableBy`	ProblemActionableBy	no	Who can act on this refusal: `user` (correct the input, sign in), `org-admin` (an administrator of the organisation), `platform` (the operators of this installation), `none` (nobody — it cannot succeed). ABSENT when not known; never guessed.
`code`	string	yes	SCREAMING_SNAKE_CASE. The machine-branchable refusal, and the value a policy rule, a retry predicate and a UI branch all name.
`detail`	string	no	Human-readable and specific to THIS occurrence.
`errors`	array of ProblemFieldIssueDto	no	Field-level issues, when the refusal is about the request body.
`instance`	string	no	URI reference identifying the occurrence — a request path, a run, a row.
`params`	map of string or number or boolean	no	Safe i18n interpolation values for the message keyed by `code` — typed and allowlisted per code by its declaration. Never server-authored copy, never internal diagnostics.
`retry`	ProblemRetry	no	What a retry can achieve: `transient` (later, unchanged), `after-change` (only after someone changes something), `never` (a defect until a fixed version ships). ABSENT when not known.
`status`	number	yes	The HTTP status code, repeated in the body so a logged document is self-contained.
`title`	string	yes	Short, human-readable, and STABLE for a given `type`. Never per-occurrence — that is `detail`.
`traceId`	string	no	W3C trace-context trace-id, present when the caller sent a `traceparent`. Omitted rather than invented when it did not.
`type`	string	yes	Absolute URI identifying the refusal TYPE. Derived one-to-one from `code`, so a caller may branch on either and can never be told two different things. Never `about:blank`.

ProblemFieldIssueDto

Property	Type	Required	Description
`code`	string	no	SCREAMING_SNAKE_CASE code of THIS issue, for a frontend i18n key. Present when the issue was minted by a declared problem.
`detail`	string	yes	Prose for logs and developers. A frontend localises from `code` + `params` instead.
`params`	map of string or number or boolean	no	Safe i18n interpolation values for this issue's `code`.
`pointer`	string	yes	RFC 6901 JSON Pointer into the request body. Root is "/".

ProblemRetry

What a retry can achieve: transient (later, unchanged), after-change (only after someone changes something), never (a defect until a fixed version ships). ABSENT when not known.

Type: one of "transient", "after-change", "never"

ProjectAgentOverrideResponseDto

Property	Type	Required	Description
`agentContentHash`	string	yes	SHA-256 of the pinned AgentRevision content.
`agentRevisionId`	string	yes	Immutable AgentRevision id pinned by this override.
`createdAt`	string (date-time)	yes
`createdBy`	string	yes
`id`	string	yes
`isActive`	boolean	yes
`mcpServicesAdd`	array of string	yes
`mcpServicesRemove`	array of string	yes
`orgId`	string	yes
`projectId`	string	yes
`systemPromptAppend`	object or null	no
`temperature`	object or null	no
`updatedAt`	string (date-time)	yes

ProjectAgentOverrideResponseDtoDataArrayEnvelope

Property	Type	Required	Description
`data`	array of ProjectAgentOverrideResponseDto	yes

ProjectAgentOverrideResponseDtoDataEnvelope

Property	Type	Required	Description
`data`	ProjectAgentOverrideResponseDto	yes

ProjectKitAuthDto

Property	Type	Required	Description
`tokenEnvVar`	string	yes	Env-var key, within the resolved wallet, holding the token.
`walletName`	string	yes	Wallet name that holds the auth token.

ProjectKitDefinitionDto

Property	Type	Required	Description
`compatibleTargets`	array of string	yes	Tech-stack tags the kit is valid for.
`defaultForTargets`	array of string	yes	Tech-stack tags for which the kit is pre-selected on a new project.
`description`	string	no	One-line kit summary.
`displayName`	string	yes	Human-readable kit name.
`enrichmentSkillSlug`	string	no	Skill slug mounted so the agent knows how to use the kit.
`kitKind`	ProjectKitKind	yes	How the kit is installed into the project.
`purpose`	ProjectKitPurpose	yes	What the kit is, from the code-producing agent’s point of view.
`slug`	string	yes	Kebab-case unique kit slug.
`source`	ProjectKitSourceDto	yes
`visibilityTier`	RegistryVisibilityTier	no	Visibility tier of this kit in the registry.

ProjectKitDefinitionDtoDataEnvelope

Property	Type	Required	Description
`data`	ProjectKitDefinitionDto	yes

ProjectKitDefinitionDtoPaginatedEnvelope

Property	Type	Required	Description
`data`	array of ProjectKitDefinitionDto	yes
`pagination`	PaginationMeta	yes

ProjectKitKind

How the kit is installed into the project.

Type: one of "npm-dependency", "shadcn-registry"

ProjectKitPurpose

What the kit is, from the code-producing agent’s point of view.

Type: one of "component-library", "generic-dependency"

ProjectKitSourceDto

Property	Type	Required	Description
`auth`	ProjectKitAuthDto	no	Wallet-backed auth for a private registry/git source.
`baseColor`	string	no	shadcn-registry source: base color palette.
`components`	array of string	no	shadcn-registry source: component names to install (e.g. ["button", "card"]).
`gitUrl`	string	no	git source: HTTPS git URL.
`iconLibrary`	string	no	shadcn-registry source: icon library (lucide \| radix).
`kind`	ProjectKitSourceKind	yes	Source discriminator.
`package`	string	no	npm-registry source: package name.
`packageName`	string	no	git source: npm package name the repository publishes.
`ref`	string	no	git source: tag / branch / commit.
`registryUrl`	string	no	npm-registry source: custom registry URL (scoped/private).
`rsc`	boolean	no	shadcn-registry source: enable React Server Components.
`scope`	string	no	npm-registry source: npm scope the registry serves.
`style`	string	no	shadcn-registry source: style preset.
`typescript`	boolean	no	shadcn-registry source: use TypeScript.
`versionRange`	string	no	npm-registry source: semver range.

ProjectKitSourceKind

Source discriminator.

Type: one of "npm-registry", "git", "shadcn-registry"

ProjectModelStrategyBindingResponseDto

Property	Type	Required
`createdAt`	string	yes
`id`	string	yes
`orgId`	string	yes
`projectId`	string	yes
`strategy`	ModelStrategyResponseDto	no
`strategyId`	string	yes
`updatedAt`	string	yes

ProjectModelStrategyBindingResponseDtoDataEnvelope

Property	Type	Required	Description
`data`	ProjectModelStrategyBindingResponseDto	yes

ProviderDeletionAffectedStrategyDto

Property	Type	Required	Description
`name`	string	yes
`strategyId`	string	yes

ProviderDeletionImpactResponseDto

Property	Type	Required	Description
`affectedOrgDefaultCount`	number	yes	Org default models (0 or 1) pointing at a removed model. Confirming the delete unsets it.
`affectedProjectIds`	array of string	yes	Projects bound to an affected strategy. Deduplicated: a project bound to two affected strategies appears once.
`affectedRuleCount`	number	yes	Every resolution RULE referencing a removed model, strategy rules and org-level rules alike. Confirming the delete removes them.
`affectedStrategies`	array of ProviderDeletionAffectedStrategyDto	yes
`canDelete`	boolean	yes	Whether the delete is permitted at all. False for system providers.
`isSystem`	boolean	yes	System providers are platform-managed and cannot be deleted; this is what `canDelete` reflects.
`modelCount`	number	yes	Models removed along with the provider.
`providerId`	string	yes
`providerName`	string	yes

ProviderDeletionImpactResponseDtoDataEnvelope

Property	Type	Required	Description
`data`	ProviderDeletionImpactResponseDto	yes

ProviderEgressUnavailableCode

PROVIDER_EGRESS_POLICY_UNAVAILABLE: the installation destination policy could not be read; nothing was sent.

Type: one of "PROVIDER_EGRESS_POLICY_UNAVAILABLE"

ProviderEgressUnavailableProblemDto

RFC 9457 problem document of an unreadable egress destination policy. code is closed.

Property	Type	Required	Description
`actionableBy`	ProblemActionableBy	no	Who can act on this refusal: `user` (correct the input, sign in), `org-admin` (an administrator of the organisation), `platform` (the operators of this installation), `none` (nobody — it cannot succeed). ABSENT when not known; never guessed.
`code`	ProviderEgressUnavailableCode	yes	`PROVIDER_EGRESS_POLICY_UNAVAILABLE`: the installation destination policy could not be read; nothing was sent.
`detail`	string	no	Human-readable and specific to THIS occurrence.
`errors`	array of ProblemFieldIssueDto	no	Field-level issues, when the refusal is about the request body.
`instance`	string	no	URI reference identifying the occurrence — a request path, a run, a row.
`params`	map of string or number or boolean	no	Safe i18n interpolation values for the message keyed by `code` — typed and allowlisted per code by its declaration. Never server-authored copy, never internal diagnostics.
`retry`	ProblemRetry	no	What a retry can achieve: `transient` (later, unchanged), `after-change` (only after someone changes something), `never` (a defect until a fixed version ships). ABSENT when not known.
`status`	number	yes	The HTTP status code, repeated in the body so a logged document is self-contained.
`title`	string	yes	Short, human-readable, and STABLE for a given `type`. Never per-occurrence — that is `detail`.
`traceId`	string	no	W3C trace-context trace-id, present when the caller sent a `traceparent`. Omitted rather than invented when it did not.
`type`	string	yes	Absolute URI identifying the refusal TYPE. Derived one-to-one from `code`, so a caller may branch on either and can never be told two different things. Never `about:blank`.

ProviderFetchBodyRefusedCode

PROVIDER_BASE_URL_NOT_ADMITTED: the provider baseUrl is outside the destinations the installation admits. PROVIDER_BASE_URL_REDIRECTED: the provider baseUrl answered with a redirect, which is never followed. REQUEST_VALIDATION_FAILED: the request body failed validation.

Type: one of "PROVIDER_BASE_URL_NOT_ADMITTED", "PROVIDER_BASE_URL_REDIRECTED", "REQUEST_VALIDATION_FAILED"

ProviderFetchBodyRefusedProblemDto

RFC 9457 problem document of a 422 from a provider probe. code is closed.

Property	Type	Required	Description
`actionableBy`	ProblemActionableBy	no	Who can act on this refusal: `user` (correct the input, sign in), `org-admin` (an administrator of the organisation), `platform` (the operators of this installation), `none` (nobody — it cannot succeed). ABSENT when not known; never guessed.
`code`	ProviderFetchBodyRefusedCode	yes	`PROVIDER_BASE_URL_NOT_ADMITTED`: the provider baseUrl is outside the destinations the installation admits. `PROVIDER_BASE_URL_REDIRECTED`: the provider baseUrl answered with a redirect, which is never followed. `REQUEST_VALIDATION_FAILED`: the request body failed validation.
`detail`	string	no	Human-readable and specific to THIS occurrence.
`errors`	array of ProblemFieldIssueDto	no	Field-level issues, when the refusal is about the request body.
`instance`	string	no	URI reference identifying the occurrence — a request path, a run, a row.
`params`	map of string or number or boolean	no	Safe i18n interpolation values for the message keyed by `code` — typed and allowlisted per code by its declaration. Never server-authored copy, never internal diagnostics.
`retry`	ProblemRetry	no	What a retry can achieve: `transient` (later, unchanged), `after-change` (only after someone changes something), `never` (a defect until a fixed version ships). ABSENT when not known.
`status`	number	yes	The HTTP status code, repeated in the body so a logged document is self-contained.
`title`	string	yes	Short, human-readable, and STABLE for a given `type`. Never per-occurrence — that is `detail`.
`traceId`	string	no	W3C trace-context trace-id, present when the caller sent a `traceparent`. Omitted rather than invented when it did not.
`type`	string	yes	Absolute URI identifying the refusal TYPE. Derived one-to-one from `code`, so a caller may branch on either and can never be told two different things. Never `about:blank`.

ProviderFetchRefusedCode

Type: one of "PROVIDER_BASE_URL_NOT_ADMITTED", "PROVIDER_BASE_URL_REDIRECTED"

ProviderFetchRefusedProblemDto

RFC 9457 problem document of a provider baseUrl egress refusal. code is closed.

Property	Type	Required	Description
`actionableBy`	ProblemActionableBy	no	Who can act on this refusal: `user` (correct the input, sign in), `org-admin` (an administrator of the organisation), `platform` (the operators of this installation), `none` (nobody — it cannot succeed). ABSENT when not known; never guessed.
`code`	ProviderFetchRefusedCode	yes	`PROVIDER_BASE_URL_NOT_ADMITTED`: the provider baseUrl is outside the destinations the installation admits. `PROVIDER_BASE_URL_REDIRECTED`: the provider baseUrl answered with a redirect, which is never followed.
`detail`	string	no	Human-readable and specific to THIS occurrence.
`errors`	array of ProblemFieldIssueDto	no	Field-level issues, when the refusal is about the request body.
`instance`	string	no	URI reference identifying the occurrence — a request path, a run, a row.
`params`	map of string or number or boolean	no	Safe i18n interpolation values for the message keyed by `code` — typed and allowlisted per code by its declaration. Never server-authored copy, never internal diagnostics.
`retry`	ProblemRetry	no	What a retry can achieve: `transient` (later, unchanged), `after-change` (only after someone changes something), `never` (a defect until a fixed version ships). ABSENT when not known.
`status`	number	yes	The HTTP status code, repeated in the body so a logged document is self-contained.
`title`	string	yes	Short, human-readable, and STABLE for a given `type`. Never per-occurrence — that is `detail`.
`traceId`	string	no	W3C trace-context trace-id, present when the caller sent a `traceparent`. Omitted rather than invented when it did not.
`type`	string	yes	Absolute URI identifying the refusal TYPE. Derived one-to-one from `code`, so a caller may branch on either and can never be told two different things. Never `about:blank`.

ProviderOwnerSpaceKind

Owner tier: system (platform-owned) or org.

Type: one of "system", "org"

ProviderProbeResponseDto

Property	Type	Required	Description
`models`	array of ProbedModelDto	yes	Models discovered at the probed endpoint (server-side, before the provider is saved).

ProviderProbeResponseDtoDataEnvelope

Property	Type	Required	Description
`data`	ProviderProbeResponseDto	yes

ProviderResponseDto

Property	Type	Required	Description
`apiType`	one of `"openai-compatible"`, `"anthropic"`, `"google"`	yes
`authType`	one of `"bearer"`, `"api-key-header"`, `"none"`	yes
`baseUrl`	string	yes
`createdAt`	string	yes
`createdBy`	string	yes
`id`	string	yes
`isActive`	boolean	yes
`isSystem`	boolean	yes
`maxConcurrentRequests`	number or null	no	Maximum concurrent requests allowed for this provider. Null = unlimited.
`models`	array of ModelResponseDto	yes
`name`	string	yes
`orgId`	string or null	yes	The owning organisation, or null for a PLATFORM-owned provider (ownerSpaceKind = system).
`ownerSpaceKind`	ProviderOwnerSpaceKind	yes	Owner tier: `system` (platform-owned) or `org`.
`slug`	string	yes

ProviderResponseDtoDataArrayEnvelope

Property	Type	Required	Description
`data`	array of ProviderResponseDto	yes

ProviderResponseDtoDataEnvelope

Property	Type	Required	Description
`data`	ProviderResponseDto	yes

ProviderTestResponseDto

Property	Type	Required	Description
`modelCount`	number	yes	Number of models returned by provider /models endpoint.

ProviderTestResponseDtoDataEnvelope

Property	Type	Required	Description
`data`	ProviderTestResponseDto	yes

ProviderWriteRefusedCode

PROVIDER_BASE_URL_NOT_ADMITTED: the provider baseUrl is outside the destinations the installation admits. REQUEST_VALIDATION_FAILED: the request body failed validation.

Type: one of "PROVIDER_BASE_URL_NOT_ADMITTED", "REQUEST_VALIDATION_FAILED"

ProviderWriteRefusedProblemDto

RFC 9457 problem document of a 422 from a provider write. code is closed.

Property	Type	Required	Description
`actionableBy`	ProblemActionableBy	no	Who can act on this refusal: `user` (correct the input, sign in), `org-admin` (an administrator of the organisation), `platform` (the operators of this installation), `none` (nobody — it cannot succeed). ABSENT when not known; never guessed.
`code`	ProviderWriteRefusedCode	yes	`PROVIDER_BASE_URL_NOT_ADMITTED`: the provider baseUrl is outside the destinations the installation admits. `REQUEST_VALIDATION_FAILED`: the request body failed validation.
`detail`	string	no	Human-readable and specific to THIS occurrence.
`errors`	array of ProblemFieldIssueDto	no	Field-level issues, when the refusal is about the request body.
`instance`	string	no	URI reference identifying the occurrence — a request path, a run, a row.
`params`	map of string or number or boolean	no	Safe i18n interpolation values for the message keyed by `code` — typed and allowlisted per code by its declaration. Never server-authored copy, never internal diagnostics.
`retry`	ProblemRetry	no	What a retry can achieve: `transient` (later, unchanged), `after-change` (only after someone changes something), `never` (a defect until a fixed version ships). ABSENT when not known.
`status`	number	yes	The HTTP status code, repeated in the body so a logged document is self-contained.
`title`	string	yes	Short, human-readable, and STABLE for a given `type`. Never per-occurrence — that is `detail`.
`traceId`	string	no	W3C trace-context trace-id, present when the caller sent a `traceparent`. Omitted rather than invented when it did not.
`type`	string	yes	Absolute URI identifying the refusal TYPE. Derived one-to-one from `code`, so a caller may branch on either and can never be told two different things. Never `about:blank`.

ProviderWriteUnavailableCode

PROVIDER_EGRESS_POLICY_UNAVAILABLE: the installation destination policy could not be read; nothing was sent. PROVIDER_TIER_POLICY_UNAVAILABLE: the organization credential policy could not be read.

Type: one of "PROVIDER_EGRESS_POLICY_UNAVAILABLE", "PROVIDER_TIER_POLICY_UNAVAILABLE"

ProviderWriteUnavailableProblemDto

RFC 9457 problem document of an unreadable policy on a provider write. code is closed.

Property	Type	Required	Description
`actionableBy`	ProblemActionableBy	no	Who can act on this refusal: `user` (correct the input, sign in), `org-admin` (an administrator of the organisation), `platform` (the operators of this installation), `none` (nobody — it cannot succeed). ABSENT when not known; never guessed.
`code`	ProviderWriteUnavailableCode	yes	`PROVIDER_EGRESS_POLICY_UNAVAILABLE`: the installation destination policy could not be read; nothing was sent. `PROVIDER_TIER_POLICY_UNAVAILABLE`: the organization credential policy could not be read.
`detail`	string	no	Human-readable and specific to THIS occurrence.
`errors`	array of ProblemFieldIssueDto	no	Field-level issues, when the refusal is about the request body.
`instance`	string	no	URI reference identifying the occurrence — a request path, a run, a row.
`params`	map of string or number or boolean	no	Safe i18n interpolation values for the message keyed by `code` — typed and allowlisted per code by its declaration. Never server-authored copy, never internal diagnostics.
`retry`	ProblemRetry	no	What a retry can achieve: `transient` (later, unchanged), `after-change` (only after someone changes something), `never` (a defect until a fixed version ships). ABSENT when not known.
`status`	number	yes	The HTTP status code, repeated in the body so a logged document is self-contained.
`title`	string	yes	Short, human-readable, and STABLE for a given `type`. Never per-occurrence — that is `detail`.
`traceId`	string	no	W3C trace-context trace-id, present when the caller sent a `traceparent`. Omitted rather than invented when it did not.
`type`	string	yes	Absolute URI identifying the refusal TYPE. Derived one-to-one from `code`, so a caller may branch on either and can never be told two different things. Never `about:blank`.

PublishAgentRevisionDto

Property	Type	Required	Description
`expectedDraftUpdatedAt`	string	yes	Current draft concurrency token. Publication fails if another author changed the draft.
`idempotencyKey`	string	yes	Caller-stable key for retry-safe publication of this exact draft.
`releaseNotes`	string	no
`sourceCommit`	string	no	Optional source commit retained as provenance only.
`sourceVersion`	string	no	Optional source-system version retained as provenance only.

RegistryVisibilityTier

Visibility tier of this kit in the registry.

Type: one of "system", "system-shareable", "org", "external"

RenderAgentRunContextRequestDto

Property	Type	Required	Description
`agentSlug`	string	yes
`changeUnitId`	string or null	no
`correlationId`	string or null	no
`deliverableSpecRef`	string or null	no
`hasRetryContext`	boolean	no	When true, the renderer seeds retry-oriented context pointers so the session knows `/workspace/deliverables/pages/` is pre-populated.
`inputArtifacts`	array of InputArtifactEntryDto	no	Files materialised into the workspace from upstream artifact-store versions. Surfaced in AGENTS.md `# Inputs` so the agent can `read` them directly without first ls'ing the slot.
`jobRunId`	string or null	no
`mounts`	array of MountEntryDto	no
`priorAttemptRef`	string or null	no
`projectId`	string	yes
`role`	string	yes	The role this invocation takes: the key of a role-capability profile some installed biome contributes. An OPEN vocabulary (contracts 0.46 deleted the closed `AGENT_RUN_ROLES` list), so the DTO validates the GRAMMAR only — the same `RESOURCE_ROLE_PATTERN` the kernel's `AgentRunRoleSchema` uses. A well-formed role no visible biome declares is refused by `RoleCapabilityRegistryService.getByRole`, which names the declared roles.
`sessionId`	string or null	no	Interactive-surface analogue of `workflowRunId`. Set by workspace-proxy when composing mount sources for an interactive session (chat / brainstorming) so `contextJson.invocation.sessionId` is populated instead of `runId`. Pipeline flows leave this null.
`workflowInputs`	object	no	Workflow job inputs (passed through from the DSL `with:` block).
`workflowRunId`	string or null	no
`workingFiles`	array of WorkingFileEntryDto	no	Working files the platform materialises before the agent's first turn. Each entry's absolute path is added to `authority.mayWriteWorkspace` so the agent may write to it; the full structured list is emitted on `context.json.workingFiles` for agent introspection.

RenderAgentRunContextResponseDto

Property	Type	Required	Description
`agentsMd`	string	yes	Full markdown body of Xema's AGENTS.md, written verbatim to the runtime-config mount (`/run/xema-runtime-config/AGENTS.md`), never to `/workspace`.
`appliedOverlays`	array of string	yes
`contextJson`	object	yes	Parsed `/workspace/context.json` payload. Serialized to JSON at write time.
`mountPlan`	RenderedMountPlanDto	yes
`renderedContextHash`	string	yes
`resolvedAgentSnapshotHash`	string	yes
`systemOverlay`	string	yes	Markdown body written verbatim to the runtime-config mount (`/run/xema-runtime-config/system-overlay.md`), never to `/workspace`. The runtime bridge plugin appends it to the LLM system prompt so the agent sees the AWP base layer + deliverable contract + authority + retry context as platform-level context, not project-level prose.
`toolAllowlist`	array of string	yes

RenderAgentRunContextResponseDtoDataEnvelope

Property	Type	Required	Description
`data`	RenderAgentRunContextResponseDto	yes

RenderedMountPlanDto

Property	Type	Required	Description
`entries`	array of MountEntryDto	yes

RescopeDto

Property	Type	Required	Description
`targetProjectId`	string	no	Project that becomes the new owner. Required when targetScope = project; rejected for any other target.
`targetScope`	ScopeSource	yes	Target owner tier. Must be a legal promotion from the row's current tier (`user → project\|org`, `project → org`). Demotion and moves into/out of `biome`/`system` are rejected with 400.

ResolveFullResponseDto

Property	Type	Required	Description
`description`	string or null	no	Agent description.
`displayName`	string	yes	Agent display name.
`executionMode`	AgentExecutionMode	yes	Execution mode (primary, subagent).
`hasProjectOverride`	boolean	yes	Whether a project-level override was applied.
`identityVersion`	number	yes	Definition identity version.
`intrinsicSubAgentPins`	map of string	yes	Definition PIN per intrinsic sub-agent slug: the content address of the agent definition that slug resolves to in this org/project scope, folded over its `extends` chain — the same value the resolver stamps as `AgentNodeDto.agentContentHash`. It reaches the warm-pool static template fingerprint, so editing an intrinsic delegate's prompt retires the pods primed against the previous one instead of serving it silently for the whole idle TTL. A slug with no active definition in scope is ABSENT from the map rather than carrying a substituted value: absent is the pre-pin identity, and an invented one would name a definition that does not exist.
`intrinsicSubAgents`	array of string	yes	Sub-agent slugs the primary may delegate to (parsed from its source manifest's `permission.task` map). These are intrinsic to the agent's design and form the FLOOR of the runtime sub-agent set — workspace/session/step bindings layer on top but never remove from this list.
`intrinsicTools`	array of string	yes	Built-in OpenCode tool keys the agent may invoke directly (parsed from its source manifest's `permission.<toolKey>` map, excluding `task`). Tools like `edit` / `write` / `read` / `bash` / `glob` / `grep` / `webfetch` / `question`. Separate from `toolProfiles` which carries the MCP-served catalog bindings.
`mcpServices`	array of string	yes	Effective MCP service keys after project override merge.
`model`	ResolvedModelStubDto	yes	Resolved LLM model.
`modelLane`	string or null	yes	The model LANE this agent routes on; null when the agent names none. An OPEN vocabulary: every key is declared by a `model-lane` contribution, so there is no closed set to publish here. `GET /work-kinds?kind=model-lane` enumerates what this organisation may use. Distinct from AgentRunRole (permissions).
`modelSelection`	ResolvedModelSelectionDto	yes	Full LLM routing record (modelId + providerSlug + opencodeProviderId + temperature + contextWindow). Carries the same data as `model` plus the worker-side `opencodeProviderId`. Decoupled from `structuralConfigHash` on purpose.
`resolvedConfigHash`	string	yes	SHA-256 hash of the canonical resolved config. Stable identity fingerprint of the full agent payload at resolve time — used by snapshot integrity checks, mcp-gateway attribution, and agent-run-context render audit. agent-session resume does NOT gate on this (resume is best-effort with a field-level drift report on the SessionResumed event); workflow runs are deterministic and rely on the snapshot identity differently.
`skills`	array of ResolvedSkillEntryDto	yes	Skills resolved (system-wide + base + pinned).
`slug`	string	yes	Agent definition slug.
`snapshotId`	string or null	no	Frozen resolved snapshot ID for this top-level resolution.
`systemPrompt`	string	yes	Merged system prompt (base + project override append).
`temperature`	number or null	no	Effective temperature. `null` defers to the engine/model default.
`toolProfiles`	array of ResolvedAgentToolProfileDto	yes	Merged tool profiles.
`workspace`	AgentWorkspaceDto or null	no	The agent's compiled workspace block — mount layout, persist paths, env and output surface — merged from its SELF-WRAP workspace manifest at biome discovery and stored verbatim on the agent. `null` for an agent that ships no self-wrap manifest: a sub-agent-only leaf is never launched as an interactive session and needs no user-data layout.

ResolveFullResponseDtoDataEnvelope

Property	Type	Required	Description
`data`	ResolveFullResponseDto	yes

ResolveModelForInvocationDto

Property	Type	Required	Description
`agentSlug`	string	no	The resolving agent — the Matrix `agent` dimension.
`modelCapability`	ModelCapability	no	A modality this invocation REQUIRES — the Matrix `modelCapability` dimension, so one rule can route every vision-needing invocation to a multimodal model without pinning a provider. It is ALSO a requirement on whatever the resolution selects: a resolved model that does not provide it is refused with 422 `MODEL_CAPABILITY_UNSUPPORTED` rather than returned, and no other model is substituted. Closed kernel set; an unknown value is a 400.
`modelId`	string	no	ABOVE-RULES concrete-model override. Pins this credentialed model and skips the strategy cascade.
`modelLane`	string	no	The model lane this invocation runs in — the Matrix `modelLane` CONTEXT dimension. Lane rules decide; it is not an override. When both this and `preferredModelLane` are supplied, this wins.
`preferredModelLane`	string	no	Per-session "smartness" preference — the Matrix `modelLane` MATCHED dimension, NOT an above-rules override. It populates the resolution context so a tier-injected managed rule can CLAMP it, and an unmapped kind matches no lane rule. Ignored when an explicit `modelLane`/`modelId` override is present. An AGENT-axis work kind: the kernel core plus every key a biome or this organisation has declared. Lower-kebab; an undeclared value is a 400 so a typo never silently lands on the no-match path — the same posture `purpose` has one field up.
`providerSlug`	string	no	Disambiguates `modelId` when one model id is credentialed under two providers. Optional partner to `modelId`.
`purpose`	string	no	Non-agent invocation purpose — the Matrix `purpose` dimension, and the MACHINE half of the contributed work-kind vocabulary. Must be a declared machine-axis work kind (`GET /work-kinds?axis=machine`); an undeclared value is a 400 so a typo never silently lands as the default rule's match path.
`skillSlug`	string	no	The active/launched skill — the Matrix `skill` dimension. Free-form: the skill vocabulary is open and owned by `skill-registry-api`, so an unknown value simply matches no rule rather than being refused.
`space`	string	yes	The Space this invocation happens in, as a canonical `xema://` SpaceRef. `xema://orgs/<orgId>` for an org-scoped invocation (a Canopy turn, a memory summarization, an org-layer session); `xema://orgs/<orgId>/projects/<projectId>` when the invocation genuinely belongs to a project. The org segment MUST match the calling context's org — it selects a Space inside the tenant the token already proved, and never widens it. Only the `project` form contributes the Matrix's `project` dimension; an org-scoped resolve carries no project dimension at all, so a `project` selector rule cannot match it.
`stageKey`	string	no	The Matrix `stage` dimension. Omit it when the call has no stage: the resolve then runs on the lane rules and the org default only, and no stage rule can match. (A caller that passes its agent slug here is treated the same: the resolver drops the dimension when the two are equal.)

ResolvedAgentKernelResponseDto

Property	Type	Required	Description
`displayName`	string	yes
`hasProjectOverride`	boolean	no
`mcpServices`	array of string	yes
`modelLane`	string or null	yes	The model LANE this agent routes on — rules select on it. An OPEN vocabulary: every key is declared by a `model-lane` contribution, and `GET /work-kinds?kind=model-lane` enumerates what this organisation may use. Distinct from AgentRunRole (permissions).
`slug`	string	yes
`source`	string	yes
`systemPrompt`	string	yes
`temperature`	number or null	no	`null` defers to the engine/model default.

ResolvedAgentKernelResponseDtoDataEnvelope

Property	Type	Required	Description
`data`	ResolvedAgentKernelResponseDto	yes

ResolvedAgentModelResponseDto

Property	Type	Required
`agentSlug`	string	yes
`modelDisplayName`	string	yes
`modelId`	string	yes
`modelRecordId`	string	yes
`providerName`	string	yes
`providerSlug`	string	yes
`source`	one of `"project"`, `"user"`, `"org"`, `"strategy"`, `"default"`	yes
`temperature`	number or null	no

ResolvedAgentModelResponseDtoDataEnvelope

Property	Type	Required	Description
`data`	ResolvedAgentModelResponseDto	yes

ResolvedAgentResponseDto

Property	Type	Required	Description
`abilities`	array of string	yes	Special-ability CapabilityRefs granted on top of the reach tier; unioned into the auto-grant manifest (so they also appear in `requiredCapabilities`). Always present (defaults to []).
`agentContentHash`	string or null	no	SHA-256 of the immutable AgentRevision content. Null only for mutable Studio drafts.
`agentRevisionId`	string or null	no	Immutable AgentRevision id selected at published resolution. Null only for mutable Studio drafts.
`capability`	AgentCapabilityLayerDto	yes	Capability envelope applied to the whole composition at runtime. Always present.
`channelVersion`	number	no
`compiledLaunchPlanHash`	string or null	no	Content address of the prepared immutable session launch plan.
`displayName`	string	yes
`durableWorkspace`	DurableWorkspaceRefDto or null	no	Resolved durable shared-workspace binding — present only when the agent declares one. When present, agent-session-api stores `workspaceRef` on `Session.workspaceRef` so the session attaches to the SHARED, workspaceRef-keyed durable workspace (multi-pod `WorkspaceSharingPolicy.Shared`) instead of a private per-session one. Absent ⇒ a private per-session workspace.
`internetEgressCeiling`	WorkerInternetEgress	no	Ceiling on whether the worker pod running this agent may open sockets to the public internet. A ceiling the agent ACCEPTS, never a grant: the effective mode is the most restrictive of the deployment, the owning Space, this value and the session, so `open` here widens nothing. RESOLVED: the agent's own declaration joined along its `extends` chain, so a child never widens its base. ABSENT when nothing in the chain declares one — there is no default.
`preparationGenerationId`	string or null	no
`reachTier`	AgentReachTier	yes	The root composition agent's reach-tier permission ceiling (Chunk 2). A composition row with no explicit tier resolves to `as-owner` (the default), so this field is always populated. agent-session-api persists it onto the session and stamps it onto `ExecutionContext.reachTier` so authorization-api applies the tier constraint inside the verdict.
`releaseSetId`	string	no
`requiredCapabilities`	array of string	yes	Chunk 3 — the deduped, sorted union of every CapabilityRef the agent needs to drive its armed tools + skills across the whole sub-agent tree (the auto-grant-from-arming manifest). Derived at resolve time from the MCP catalog (tools) + skill frontmatter (skills). Empty when nothing armed declares a capability — the default-safe no-op. The capability-analysis endpoint returns the same union plus the unresolved-ref diagnostics.
`resourceId`	string	yes	Stable Agent AuthoredResource identity selected by resolution.
`root`	AgentNodeDto	yes	Resolved root node — Agent fully revision-pinned, recursive children resolved.
`runtimeTemplateHash`	string or null	no	Hash of the prepared plan RuntimeTemplateKey.
`scope`	ScopeSource	yes
`sessionLifecycle`	AgentSessionLifecycle or null	no	The resolved agent's declared default session lifecycle (`ephemeral`\|`persistent`), or null when it declares no preference. agent-session-api inherits this as the session `lifecyclePolicy` unless the launch passes an explicit override; null ⇒ agent-session applies its own `ephemeral` default last.
`slug`	string	yes
`standingSession`	StandingSessionRefDto or null	no	Resolved standing-session pin — present only when the agent declares one. When present, agent-session-api does a `(orgId, projectId, pinKey)` get-or-create at `POST /sessions` and ADOPTS a non-terminal existing session instead of minting a fresh one. Absent ⇒ a fresh session per launch.
`version`	string	yes	Server-generated immutable revision number.
`workspace`	AgentWorkspaceDto	no	Resolved workspace-layout config — present when the source composition declared a workspace block. The `${input.x}` interpolation in env values is already resolved here. The workspace pipeline projects this onto the run's WorkspaceSpec.
`workspaceSharing`	WorkspaceSharingPolicy	yes	Resolved workspace-sharing policy — whether a session launch of this agent may POOL onto a compatible running pod as a new thread (`shareable`) or must always run in its own isolated worker (`isolated`). Always populated: a row with no stored value resolves to `isolated` (WORKSPACE_SHARING_DEFAULT). agent-session-api reads this to gate session-thread pooling at launch.

ResolvedAgentResponseDtoDataEnvelope

Property	Type	Required	Description
`data`	ResolvedAgentResponseDto	yes

ResolvedAgentToolProfileDto

Property	Type	Required	Description
`allowedTools`	array of string	yes	Explicitly allowed tools.
`catalogId`	string or null	no	Catalog ID for the tool source.
`isOptional`	boolean	yes	Whether this tool profile is optional.
`mcpServiceKey`	string	yes	MCP service key.
`toolGroups`	array of string	yes	Tool groups enabled for this service.

ResolvedConfigSnapshotResponseDto

Property	Type	Required	Description
`agentSlug`	string	yes	Agent slug captured by this snapshot.
`canonicalizationVersion`	number	yes	Canonicalization version.
`createdAt`	string (date-time)	yes	Snapshot creation timestamp.
`id`	string	yes	Snapshot ID.
`invocationId`	string	yes	Invocation ID associated with this snapshot.
`payload`	object or null	no	Inline snapshot payload stored in the database.
`payloadObjectKey`	object or null	no	External payload object key when the snapshot is offloaded from the database.
`resolvedConfigHash`	string	yes	Resolved config hash captured by this snapshot.
`snapshotSchemaVersion`	number	yes	Snapshot schema version.

ResolvedConfigSnapshotResponseDtoDataEnvelope

Property	Type	Required	Description
`data`	ResolvedConfigSnapshotResponseDto	yes

ResolvedDelegatedApertureDto

Property	Type	Required	Description
`agentKey`	string	yes
`agentRef`	DelegatedAgentRefDto	yes
`authoredPath`	array of DelegatedAgentSubjectDto	yes
`constraints`	DelegatedApertureConstraintsDto	yes
`delegateAgentKeys`	array of string	yes	Every entry this authority may delegate to (all routes, not only the canonical parent), sorted.
`depth`	number	yes
`parentAgentKey`	string or null	yes
`reachTier`	one of `"sandbox"`, `"as-owner"`, `"connected"`, `"full"`	yes

ResolvedEmbeddingCommitmentDto

Property	Type	Required	Description
`commitmentId`	string	yes
`dimension`	number	yes
`migration`	MigrationTargetDto or null	no	Present only when state = migrating. Consumers must dual-write to both the current model/dimension and this target.
`modelId`	string	yes
`providerApiType`	string	yes
`providerBaseUrl`	string	yes
`providerId`	string	yes
`providerIsSystem`	boolean	yes
`providerName`	string	yes
`providerSlug`	string	yes
`state`	one of `"proposed"`, `"active"`, `"migrating"`	yes

ResolvedEmbeddingCommitmentDtoDataEnvelope

Property	Type	Required	Description
`data`	ResolvedEmbeddingCommitmentDto	yes

ResolvedModelDecisionDto

Property	Type	Required	Description
`matchedDimensions`	array of string	yes	Dimensions the winning rule matched on.
`matchedRuleId`	string or null	no	Id of the rule that won; null when the implicit default fallback was used.
`model`	ResolvedModelRefDto	yes	The resolved model reference.
`overridden`	boolean	yes	True when a run/realtime override pre-empted the matrix.
`reason`	string	yes	Human-readable win reason.

ResolvedModelRefDto

Property	Type	Required	Description
`kind`	one of `"concrete"`	yes	Always `concrete`: a rule targets a model.
`modelId`	string or null	no	Concrete model id.
`providerSlug`	string or null	no	Provider slug pinned by the rule.
`reasoningEffort`	string or null	no
`temperature`	number or null	no

ResolvedModelSelectionDto

Property	Type	Required	Description
`capabilities`	array of string	no	Model capabilities (e.g. vision).
`contextWindow`	number or null	no	Context window (tokens).
`modelId`	string or null	no
`opencodeProviderId`	string or null	no	OpenCode provider ID (worker-side routing).
`providerFamily`	string	no	llm-gateway provider family (open vocabulary). OMITTED when unknown; an absent value means unknown.
`providerSlug`	string or null	no
`source`	one of `"matrix"`, `"override"`, `"org_default"`, `"unresolved"`	yes	Source of the model resolution.
`temperature`	number or null	no	`null` defers to the engine/model default.

ResolvedModelStubDto

Property	Type	Required	Description
`modelId`	string or null	no	Resolved model ID (null when unresolved).
`providerSlug`	string or null	no	Provider slug (null when unresolved).
`source`	one of `"matrix"`, `"override"`, `"org_default"`, `"unresolved"`	yes	Source of the model resolution.
`temperature`	number or null	no	Temperature for the model. `null` defers to the engine/model default.

ResolvedPhaseConfigDto

Property	Type	Required	Description
`contextWindow`	number or null	no	The model's TOTAL token budget (prompt + reasoning + answer). Never an input cap.
`decision`	ResolvedModelDecisionDto or null	no	Model Resolution Matrix win-reason trace — which rule won and why. Null only for the persisted per-agent override path.
`llmModelId`	string	yes	The registry model record id (`LlmModel.id`) the resolution selected — the id a rule or the org default targets.
`maxInputTokens`	number or null	no	A declared cap on the PROMPT alone, independent of the window. Null = none declared; only the window bounds the prompt.
`maxOutputTokens`	number or null	no	Declared ceiling on one answer. Null = UNKNOWN — never "unlimited", never "the rest of the window".
`modelDisplayName`	string	yes
`modelId`	string	yes	The provider's own model id.
`providerApiType`	string	yes
`providerBaseUrl`	string	yes
`providerFamily`	string	no	llm-gateway provider family of the provider (open vocabulary). OMITTED when unknown; an absent value means unknown.
`providerId`	string	yes
`providerName`	string	yes
`reasoningEffort`	string or null	no	Reasoning effort the resolution path selected, forwarded verbatim to the provider. `null` defers to the model default — never a substituted value.
`retirement`	ModelRetirementSubstitutionDto	no	Present only when the picked model was retired and its successor was served instead; `decision` still names the pick.
`source`	string	yes
`stageKey`	string	yes
`temperature`	number or null	no	`null` defers to the engine/model default.

ResolvedPhaseConfigDtoDataEnvelope

Property	Type	Required	Description
`data`	ResolvedPhaseConfigDto	yes

ResolvedSkillEntryDto

Property	Type	Required	Description
`displayName`	string or null	no	Display name.
`key`	string	yes	Skill key/slug.
`source`	one of `"system"`, `"base"`	yes	How this skill was selected (resolution layer).

ResolvedStageModelConfigDto

Property	Type	Required	Description
`apiType`	string	yes	Wire protocol the provider speaks, as stored on the provider row.
`baseUrl`	string	yes	Provider base URL the invocation would call.
`decision`	ResolvedModelDecisionDto or null	no	Model Resolution Matrix win-reason trace — which rule won and why. Null only for the persisted per-agent override path.
`modelId`	string	yes	Provider-native model id this stage currently routes to.
`providerId`	string	yes	Registry id of the provider serving the model.
`providerName`	string	yes	Provider display name.
`providerSlug`	string	yes	Provider slug.
`reasoningEffort`	string or null	no	Reasoning effort this stage currently resolves to. `null` defers to the model default.
`retirement`	ModelRetirementSubstitutionDto	no	Present only when the stage's picked model was retired and its successor is served instead; `decision` still names the pick.
`source`	ModelResolutionSource	yes	Which resolution path produced this selection.
`temperature`	number or null	no	`null` defers to the engine/model default.

ResolvedToolProfileDto

Property	Type	Required	Description
`allowedTools`	array of string	yes
`catalogId`	object or null	no
`isOptional`	boolean	yes
`mcpServiceKey`	string	yes
`source`	ScopeSource	yes	Owner tier that resolved this profile.
`toolGroups`	array of string	yes

ResolvedToolProfileDtoDataArrayEnvelope

Property	Type	Required	Description
`data`	array of ResolvedToolProfileDto	yes

RestoreAgentRevisionDto

Property	Type	Required	Description
`expectedDraftUpdatedAt`	string	yes	Current draft concurrency token. Restore fails if another author changed the draft.

RetirementModelRefDto

Property	Type	Required	Description
`modelId`	string	yes	The provider's model id.
`recordId`	string	yes	The model record id (`LlmModel.id`).

RevisionPinDto

Property	Type	Required	Description
`holder`	RevisionPinHolder	yes	What may be DONE with this pin. Only `mutable` pins are movable; `immutable` needs the holder republished, and `provenance` records what happened and is never touched.
`holderId`	string	yes
`holderKind`	string	yes	Which kind of consumer holds it, e.g. `project-agent-override`.
`label`	string or null	yes
`resourceId`	string	yes	The Agent identity that owns the revision.
`revisionId`	string	yes

RevisionPinDtoDataArrayEnvelope

Property	Type	Required	Description
`data`	array of RevisionPinDto	yes

RevisionPinHolder

What may be DONE with this pin. Only mutable pins are movable; immutable needs the holder republished, and provenance records what happened and is never touched.

Type: one of "mutable", "immutable", "provenance"

RevisionPinMovePlanDto

Property	Type	Required	Description
`immovable`	array of RevisionPinDto	yes	Pins that record what DID run. Never moved: moving one would not re-point a consumer, it would rewrite history.
`move`	array of RevisionPinDto	yes	Pins that were moved.
`republishRequired`	array of RevisionPinDto	yes	Pins embedded inside a published revision. Republish the holder to change them — rewriting one would mutate an immutable revision.

RevisionPinMovePlanDtoDataEnvelope

Property	Type	Required	Description
`data`	RevisionPinMovePlanDto	yes

RoleCapabilityProfileDto

Property	Type	Required	Description
`mayEditRepos`	one of `"never"`, `"when-granted"`	yes	Whether the role may edit repositories
`mayWriteKnowledgeBase`	boolean	yes	Knowledge-base write capability
`role`	string	yes	AgentRunRole slug — kebab-case
`surface`	one of `"pipeline-run"`, `"agent-session"`, `"gate-review"`	yes	Where this role may run
`writePolicy`	one of `"read-only"`, `"own-pages-only"`, `"full-write"`	yes	Workspace write policy

RoleCapabilityProfileDtoDataEnvelope

Property	Type	Required	Description
`data`	RoleCapabilityProfileDto	yes

RoleCapabilityProfileDtoPaginatedEnvelope

Property	Type	Required	Description
`data`	array of RoleCapabilityProfileDto	yes
`pagination`	PaginationMeta	yes

RollbackAgentRevisionDto

Property	Type	Required	Description
`expectedChannelVersion`	number	yes
`priorReleaseSetId`	string	yes	Owner-fenced immutable ReleaseSet whose stable state should be replayed.

ScopeSource

Type: one of "system", "biome", "org", "project", "user"

SetModelRetirementDto

Property	Type	Required	Description
`retiresAt`	string (date-time)	yes	The retirement instant (ISO 8601). From this instant on the model is never served as picked.
`successorModelId`	string or null	no	The model record id (`LlmModel.id`) resolutions use instead from `retiresAt`. Must be served to the organisation, must not be the model itself, and its own successor chain must reach a model not retired at `retiresAt` without cycling. Omit or null: the model is refused from `retiresAt`.

SetOrgModelDefaultDto

Property	Type	Required	Description
`modelId`	string	yes	The registry model record id (`LlmModel.id`) the organisation routes to when nothing more specific applies. Must be served by an active provider.

StandingSessionRefDto

Property	Type	Required	Description
`pinKey`	string	yes	Stable key identifying the standing session for a purpose/biome, e.g. `{biomeId}:{purpose}`. Two launches of the same agent within the same `(orgId, projectId)` that resolve this key reuse ONE session row.

StartEmbeddingMigrationDto

Property	Type	Required	Description
`acknowledgeReprocessing`	boolean	yes	The user must explicitly tick a checkbox acknowledging that all existing vectors will be reprocessed. Sent as `true`; the backend rejects the migration if false.
`targetModelId`	string	yes	Target embedding model id.
`targetProviderId`	string	yes	Target provider id.
`typedOrgSlug`	string	yes	Required confirmation: the org slug typed by the user. The backend rejects the migration if this does not match the actual org slug.

StrategyDeletionAffectedProjectDto

Property	Type	Required	Description
`projectId`	string	yes	Project whose binding cascade-deletes.

StrategyDeletionImpactResponseDto

Property	Type	Required	Description
`affectedProjects`	array of StrategyDeletionAffectedProjectDto	yes
`boundProjectCount`	number	yes	Projects explicitly bound to this strategy. Their bindings cascade-delete.
`displayName`	string	yes
`strategyId`	string	yes

StrategyDeletionImpactResponseDtoDataEnvelope

Property	Type	Required	Description
`data`	StrategyDeletionImpactResponseDto	yes

SurfaceKindDto

Property	Type	Required	Description
`biomeId`	string	no	Present only for biome-contributed kinds.
`displayName`	string	yes	Human-readable display name.
`kind`	string	yes	Fully-namespaced kind slug the FE registry keys by. A first-party kind is its bare enum value (e.g. `markdown`); a biome kind is `<biomeId>:<kind>`.
`payloadSchema`	object	yes	JSON Schema for the surface payload (passed through verbatim; the FE derives its generic renderer from it).
`source`	SurfaceKindSource	yes	Provenance of this kind.
`summary`	string	yes	What the surface shows and when to open it.
`version`	number	yes	Monotonic payload schema version.

SurfaceKindListDto

Property	Type	Required	Description
`surfaceKinds`	array of SurfaceKindDto	yes

SurfaceKindListDtoDataEnvelope

Property	Type	Required	Description
`data`	SurfaceKindListDto	yes

SurfaceKindSource

Provenance of this kind.

Type: one of "first-party", "biome"

SyncModelsResponseDto

Property	Type	Required	Description
`discoveredCount`	number	yes	Total number of models discovered from provider /models.
`existingCount`	number	yes	Number of discovered models already registered in llm-model table.
`existingModelIds`	array of string	yes	Model IDs that were already present before this sync.
`modelIds`	array of string	yes	Model IDs newly inserted during this sync.
`skippedCount`	number	yes	Number of discovered models skipped because required metadata was missing.
`skippedModelIds`	array of string	yes	Model IDs skipped due to missing required metadata (e.g. contextWindow).
`syncedCount`	number	yes	Number of new models inserted into llm-model table.

SyncModelsResponseDtoDataEnvelope

Property	Type	Required	Description
`data`	SyncModelsResponseDto	yes

ToolProfileResponseDto

Property	Type	Required	Description
`agentSlug`	string	yes
`allowedTools`	array of string	yes
`biomeId`	string or null	no	Biome id when source='biome', otherwise null.
`catalogId`	string or null	no
`createdAt`	string (date-time)	yes
`forkedFrom`	string or null	no	Origin reference (`<source>:<biomeId\|"">:<key>`) when this row was forked from a biome/system tool profile.
`id`	string	yes
`isActive`	boolean	yes
`isOptional`	boolean	yes
`mcpServiceKey`	string	yes
`orgId`	string or null	no
`ownerUserId`	string or null	no	Owning user id when source='user', otherwise null.
`projectId`	string or null	no
`source`	ScopeSource	yes	Owner tier — the kernel's `SpaceKind`, narrowed to the five tiers this registry owns. The wire schema keeps the name `ScopeSource`. Resolution precedence at read time: user > project > org > biome > system.
`toolGroups`	array of string	yes
`updatedAt`	string (date-time)	yes
`version`	number	yes

ToolProfileResponseDtoDataArrayEnvelope

Property	Type	Required	Description
`data`	array of ToolProfileResponseDto	yes

ToolProfileResponseDtoDataEnvelope

Property	Type	Required	Description
`data`	ToolProfileResponseDto	yes

ToolsetKey

Type: one of "read", "write", "exec", "web", "mcp", "delegate"

UnresolvedLaneDto

Property	Type	Required	Description
`agentSlug`	string	yes	Slug of the agent whose model cannot be resolved.
`message`	string	yes	Human-readable statement of what is missing.
`modelLane`	string or null	yes	The model lane the agent asks for; null when it asks for none (only lane-less rules and the organisation default can answer it).
`reason`	ModelResolutionErrorCode	yes	The code a launch of this agent is refused with — the same value the resolve doors put on the refusal.

UnresolvedToolDto

Property	Type	Required
`providerKind`	string	yes
`resourceId`	string	yes
`toolName`	string	no

UpdateAgentDraftDto

Property	Type	Required	Description
`abilities`	array of string	no	Special-ability CapabilityRefs granted on top of the reach tier; unioned into the auto-grant manifest.
`capability`	AgentCapabilityLayerDto	no
`description`	string	no
`displayName`	string	no
`durableWorkspace`	DurableWorkspaceRefDto or null	no	Durable shared-workspace binding. Omitted leaves the stored value unchanged; set `{ workspaceRef }` to attach every session this agent launches to ONE durable shared workspace; pass `null` to clear it (the agent reverts to a private per-session workspace).
`expectedDraftUpdatedAt`	string	yes	Current draft concurrency token. The update fails on stale writes.
`internetEgressCeiling`	WorkerInternetEgress or null	no	Ceiling on whether the worker pod running this agent may open sockets to the public internet. A ceiling the agent ACCEPTS, never a grant: the effective mode is the most restrictive of the deployment, the owning Space, this value and the session, so `open` here widens nothing. Omitted leaves the stored value unchanged; `null` CLEARS it, so the agent states nothing again.
`reachTier`	AgentReachTier	no	The agent's permission ceiling. Omitted leaves the stored value unchanged; a NULL stored value reads as `as-owner`.
`root`	AgentNodeDto	no
`sessionLifecycle`	AgentSessionLifecycle	no	The agent's default session lifecycle (`ephemeral`\|`persistent`). Omitted leaves the stored value unchanged; a NULL stored value means the agent declares no preference.
`standingSession`	StandingSessionRefDto or null	no	Standing-session pin. Omitted leaves the stored value unchanged; set `{ pinKey }` to make every launch reuse ONE session keyed by `pinKey`; pass `null` to clear the pin (the agent stops being a standing agent).
`workspace`	AgentWorkspaceDto	no	Authored workspace-layout config — persist-paths + per-slot overrides, env vars, output surface.
`workspaceSharing`	WorkspaceSharingPolicy	no	Whether a session launch may POOL onto a compatible running pod (`shareable`) or must always run isolated (`isolated`). Omitted leaves the stored value unchanged; the column default is `isolated`.

UpdateAgentKernelDto

Property	Type	Required	Description
`avatarSeed`	string	no
`avatarStyle`	string	no
`description`	string	no
`displayName`	string	no
`isActive`	boolean	no
`mcpServices`	array of string	no
`modelLane`	string	no	The model LANE this agent routes on — rules select on it. An OPEN vocabulary: every key is declared by a `model-lane` contribution, and `GET /work-kinds?kind=model-lane` enumerates what this organisation may use. Distinct from AgentRunRole (permissions).
`skills`	array of string	no
`systemPrompt`	string	no
`temperature`	number	no

UpdateCodeStandardDto

Property	Type	Required
`content`	string	no
`isActive`	boolean	no
`name`	string	no

UpdateModelDto

Property	Type	Required	Description
`capabilities`	array of string	no	Model capabilities
`contextWindow`	number	no	Context window (max input tokens). Recorded with source `operator`; the provider sync never overwrites an operator value. Not nullable — a model always has a context window. Omit to leave it unchanged.
`displayName`	string	no	Human-friendly display name
`embeddingDimension`	number	no	Embedding vector dimension (for embedding models only)
`isActive`	boolean	no	Whether the model is active and available for selection
`maxInputTokens`	number or null	no	Max prompt (input) tokens, when the service caps input below the context window. A positive integer is recorded with source `operator`. `null` clears the fact to unknown (value and source). Omit to leave it unchanged.
`maxOutputTokens`	number or null	no	Max generation/output tokens. A positive integer is recorded with source `operator` and the provider sync never overwrites it. `null` clears the fact to unknown (value and source), so the next provider sync may fill it again. Omit to leave it unchanged.
`modalities`	array of ModelModality	no	Closed-enum INPUT modalities the model accepts — an operator declaration recorded with source `operator`, which no sync overwrites. An empty array clears the fact to unknown so the next sync may fill it. Omit to leave it unchanged.
`supportsAudioInput`	boolean or null	no	The model accepts audio inputs. `true`/`false` is recorded with source `operator`, which no sync overwrites. `null` clears the fact to unknown so the next sync may fill it. Omit to leave it unchanged.
`supportsJsonMode`	boolean or null	no	The model supports structured JSON output. `true`/`false` is recorded with source `operator`, which no sync overwrites. `null` clears the fact to unknown so the next sync may fill it. Omit to leave it unchanged.
`supportsPdfInput`	boolean or null	no	The model accepts PDF inputs. `true`/`false` is recorded with source `operator`, which no sync overwrites. `null` clears the fact to unknown so the next sync may fill it. Omit to leave it unchanged.
`supportsPromptCaching`	boolean or null	no	The model supports prompt caching. `true`/`false` is recorded with source `operator`, which no sync overwrites. `null` clears the fact to unknown so the next sync may fill it. Omit to leave it unchanged.
`supportsReasoning`	boolean or null	no	The model performs explicit reasoning. `true`/`false` is recorded with source `operator`, which no sync overwrites. `null` clears the fact to unknown so the next sync may fill it. Omit to leave it unchanged.
`supportsTools`	boolean or null	no	The model exposes function/tool calling. `true`/`false` is recorded with source `operator`, which no sync overwrites. `null` clears the fact to unknown so the next sync may fill it. Omit to leave it unchanged.
`supportsVision`	boolean or null	no	The model accepts image inputs. `true`/`false` is recorded with source `operator`, which no sync overwrites. `null` clears the fact to unknown so the next sync may fill it. Omit to leave it unchanged.
`supportsWebSearch`	boolean or null	no	The model supports built-in web search. `true`/`false` is recorded with source `operator`, which no sync overwrites. `null` clears the fact to unknown so the next sync may fill it. Omit to leave it unchanged.

UpdateModelResolutionRuleDto

Property	Type	Required
`priority`	number	no
`selector`	ModelResolutionSelectorDto	no
`targetModelId`	string	no
`targetReasoningEffort`	string	no
`targetTemperature`	number	no

UpdateModelStrategyDto

Property	Type	Required	Description
`description`	string or null	no
`displayName`	string	no

UpdateProjectAgentOverrideDto

Property	Type	Required
`isActive`	boolean	no
`mcpServicesAdd`	array of string	no
`mcpServicesRemove`	array of string	no
`systemPromptAppend`	string	no
`temperature`	number	no

UpdateProviderDto

Property	Type	Required	Description
`apiKey`	string	no	Optional enrolment transport: when supplied, the key is sealed in the credential plane and becomes the organisation's default credential (org_default pointer) for this provider's key; the provider row never stores it. Omit it when members use their own keys.
`apiType`	one of `"openai-compatible"`, `"anthropic"`, `"google"`	no
`authType`	one of `"bearer"`, `"api-key-header"`, `"none"`	no
`baseUrl`	string	no
`isActive`	boolean	no
`maxConcurrentRequests`	number or null	no	Maximum concurrent requests. Set to null to remove the limit.
`name`	string	no
`slug`	string	no

UpsertAgentModelOverrideDto

Property	Type	Required	Description
`agentSlug`	string	yes	Agent slug to override the model for.
`modelId`	string	yes	LLM model ID to assign to this agent.
`scope`	one of `"org"`, `"user"`, `"project"`	yes	Override scope: "org" (org-wide default), "user" (personal preference), "project" (project-specific).
`scopeId`	string	yes	Scope identifier: orgId for "org", userId for "user", projectId for "project".
`temperature`	number	no	Optional temperature override (0.0 - 2.0).

UpsertEmbeddingCommitmentDto

Property	Type	Required	Description
`modelId`	string	yes	Embedding model id (e.g. "text-embedding-3-small"). The model must be registered in the org's providers and have the "embeddings" capability.
`providerId`	string	yes	Provider id that hosts the model.

UpsertModelStrategyLaneRuleDto

Property	Type	Required	Description
`priority`	number	no	Tie-break hint when two rules match the same number of dimensions and the fixed dimension-priority ladder does not separate them. Higher wins. Defaults to 0.
`targetModelId`	string	yes	Registry model record id (LlmModel.id) the rule routes to. It must be an active model on an active provider the organisation or the platform owns; otherwise the write is refused with 422 MODEL_NOT_SERVED.
`targetReasoningEffort`	string	no	Optional reasoning effort carried by the resolved model, forwarded verbatim to the provider. OPEN vocabulary owned by the provider (e.g. minimal/low/medium/high, or low/medium/high/xhigh/max) — not an enum here, so a new provider spelling is not refused at the door.
`targetTemperature`	number	no	Optional temperature override carried by the resolved model.

UpsertModelStrategyRulesDto

Property	Type	Required	Description
`expectedVersion`	number	no	The `version` read from this strategy. When sent, a stale value is refused with 409 `MODEL_STRATEGY_VERSION_MISMATCH`, which carries `currentVersion` to retry with. Send it: it is optional only for one transition and will become REQUIRED — without it this route replaces the whole rule set unconditionally, so a concurrent edit to a DIFFERENT lane is silently discarded. To change one lane, use PUT /model-strategies/:id/rules/lane/:lane, which needs no version.
`rules`	array of CreateModelResolutionRuleDto	yes	The strategy's rules (replaces the whole rule set). Each rule has the same shape and the same validation as an org-level model resolution rule.

UpsertOrgLaneRuleDto

Property	Type	Required	Description
`priority`	number	no	Tie-break hint when two rules match the same number of dimensions and the fixed dimension-priority ladder does not separate them. Higher wins. Defaults to 0.
`targetModelId`	string	yes	Registry model record id (LlmModel.id) the rule routes to. It must be an active model on an active provider the organisation or the platform owns; otherwise the write is refused with 422 MODEL_NOT_SERVED.
`targetReasoningEffort`	string	no	Optional reasoning effort carried by the resolved model, forwarded verbatim to the provider. OPEN vocabulary owned by the provider (e.g. minimal/low/medium/high, or low/medium/high/xhigh/max) — not an enum here, so a new provider spelling is not refused at the door.
`targetTemperature`	number	no	Optional temperature override carried by the resolved model.

UpsertToolProfileDto

Property	Type	Required	Description
`allowedTools`	array of string	no	Explicitly allowed tool names.
`catalogId`	string	no	Catalog identifier for tool discovery.
`expectedVersion`	number	yes	Optimistic concurrency version. Use 0 for create.
`isOptional`	boolean	no	Whether this MCP service is optional for the agent.
`toolGroups`	array of string	no	Tool group identifiers.

WidgetKindDto

Property	Type	Required	Description
`biomeId`	string	no	Present only for biome-contributed kinds.
`displayName`	string	yes	Human-readable display name.
`kind`	string	yes	Fully-namespaced kind slug the FE registry keys by. A first-party kind is its bare enum value (e.g. `plan`); a biome kind is `<biomeId>:<kind>`.
`payloadSchema`	object	yes	JSON Schema for the widget payload (passed through verbatim).
`source`	WidgetKindSource	yes	Provenance of this kind.
`summary`	string	yes	What the widget shows and when to use it.
`version`	number	yes	Monotonic payload schema version.

WidgetKindListDto

Property	Type	Required	Description
`widgetKinds`	array of WidgetKindDto	yes

WidgetKindListDtoDataEnvelope

Property	Type	Required	Description
`data`	WidgetKindListDto	yes

WidgetKindSource

Provenance of this kind.

Type: one of "first-party", "biome"

WorkerInternetEgress

Ceiling on whether the worker pod running this agent may open sockets to the public internet. A ceiling the agent ACCEPTS, never a grant: the effective mode is the most restrictive of the deployment, the owning Space, this value and the session, so open here widens nothing. RESOLVED: the agent's own declaration joined along its extends chain, so a child never widens its base. ABSENT when nothing in the chain declares one — there is no default.

Type: one of "open", "governed"

WorkingFileEntryDto

Property	Type	Required
`format`	string	yes
`path`	string	yes
`slug`	string	yes
`sourceKind`	string	yes
`sourceRef`	map of string	yes
`syncDirection`	string	yes

WorkspaceSharingPolicy

Resolved workspace-sharing policy — whether a session launch of this agent may POOL onto a compatible running pod as a new thread (shareable) or must always run in its own isolated worker (isolated). Always populated: a row with no stored value resolves to isolated (WORKSPACE_SHARING_DEFAULT). agent-session-api reads this to gate session-thread pooling at launch.

Type: one of "isolated", "shareable", "shared"