Platform reference

The platform reference describes the system layer of Xema. It defines the typed objects, the capability-mediated invocation plane, the execution environments, and the installable software bundles (biomes) that humans, agents, workflows, and external apps all share.

This section is the public reference for the Xema platform — for org admins choosing what to install, for biome authors building extensions, and for integrators embedding Xema into their own products.

PageWhat it covers
OverviewWhat Xema is, the conceptual architecture layers, and the four Agent primitives
ObjectsXemaObject, the typed universe, XVFS read path
CapabilitiesCapability refs, the gateway, two-stage permission model
ConnectorsTyped integration points, named multi-account connections, wallet credentials, OAuth flow, first-party connectors
Conversation channelsTelegram and Slack to durable agent sessions: channel accounts, coalesced replies, the sandboxed default agent, resending uncertain replies
BiomesThe installable software bundle, the seven-state lifecycle, the install flow
Execution EnvironmentsThe nine built-in environments, the reach ceiling, environment grants
SpacesThe seven-level SpaceRef hierarchy, URI grammar, data classification, re-scoping
Execution ContextsThe per-invocation envelope; what flows where
PolicyPolicyDecision shape, the closed obligation set, route hints, credential selection, approval flow
Permissions & AccessThe whole access model — capabilities, grants, roles, teams, agent reach tiers, auto-grant
ProfilesNamed grant bundles attached to subjects — planned, not implemented
RunnersThe runner kinds, execution targets, enrollment ceilings, attestation, push vs pull, the selector
ControllersReconciliation loops, desired vs observed state, error categories
Service RegistryService discovery + @InjectService(name)
Hook planeAdmit, transform or defer an operation before it commits
Process topologyOwn process or composed process, and what it means for biome code
Distributed initDesign, not yet available: kernel join, runtime registration, permissions as data
DecisionsThe one primitive for a human choice: approvals, consent, questions
Governed web accessThe audited door to the public web: page fetch, search, reach-tier gating, the untrusted-content envelope
MCP and CapabilitiesThe six meta-tools + external MCP federation
Developer Annotations@XemaResource / @XemaRoute + convention inference
CLIThe global xema CLI — install, commands, workflows
Xema-as-CodeDeclarative provisioning — xema.yaml, the Terraform provider, the managedBy ownership model
Public exposure and CORSWhich public API routes the gateway serves on which host, and which browser origins may call them
SkillsSkill bundles, Space ownership, authoring and slash commands
Agent CompositionRecursive agent tree, agent lifecycle, Model Resolution Matrix
ShellThe unified command surface, WebSocket protocol, sandbox terminal
CanopyThe universal agent: conversations and threads, Tier A and Tier B, autonomy levels, consent and standing grants
StoreThe four Store capabilities, listing lifecycle, submission and review flow (the Store is a catalogue; installs go through biome lifecycle)
EditionsThe six editions that ship, what each can and cannot do, the three installation classes, and how to choose one
VersioningDraft vs published, rollback as a pointer flip, lockfile shape, capability auto-bump
AppsApp model, delegated sessions and the signing key ring, audience policies, ingress caps, embed snippet
Design System BuilderAgent-authored design systems: components, tokens, docs, pipeline handoff and publication
Document BuddyCo-authoring A4 Markdown and rich documents with an agent: tracked changes, templates, design systems, the Presenter Coach
TopologyThe project manifest as a live map of services, resources, integrations and history
Resource GovernanceResource inventory, usage ingest, admission, cost inputs and the release plane
SDK / Getting StartedAuthor your first biome with the xema CLI
Manifest ReferenceGenerated, field-by-field contract for the current xema-biome.json schema
SDK / ContributionsAuthoring the Contribution Protocol — *.contribution.json files
SDK / Lifecycle HooksonInstall, onUninstall, onUpgrade, onEnable, onDisable — declared, not yet invoked
SDK / Backend I shipComponent-based service, worker, job, adapter, content, and web artifacts
SDK / StorageDeclared collections, isolation, uninstallPolicy
SDK / PublishingThe four publishing transitions, bundle format, signing intent
SDK / UI I contributeHostExtensionSlots, route contributions, nav registry
SDK / Events I consumePer-service event consumers, CloudEvents envelope
SDK / TestingManifest validation, capability-stub tests, lifecycle hook tests

Getting Started

Read in order:

  1. Overview — orient yourself in the conceptual layers and the four-concept runtime model.
  2. Objects — see how everything in Xema is a typed XemaObject.
  3. Spaces — where data lives and who owns it.
  4. Execution Environments — the trust profiles capability calls run under.
  5. Execution Contexts — the per-invocation envelope.
  6. Policy — how every invocation is decided.
  7. Permissions & Access — the whole access model + the agent reach-tier ceiling.
  8. Capabilities — the invocation surface every call uses.
  9. Biomes — the lifecycle every installable bundle moves through.
  10. Skills — teach agents domain knowledge with folder bundles.
  11. Agent Composition — build multi-agent pipelines with the Model Resolution Matrix.
  12. Shell — the command surface humans and agents share.
  13. CLI — install and use the xema binary.
  14. Xema-as-Code — provision the platform declaratively with xema.yaml or Terraform.
  15. SDK / Getting Started — author your first biome.
  16. Manifest Reference — reference for every current manifest field.