Decision Terrain Developer Documentation

Developer docs cover documented public integration surfaces. The product UI also has many authenticated console routes, but those are not public APIs unless documented here.

Start here

• Developer Basics: authentication, base URLs, environments, and conventions.

• API quickstart: create an agent, send a message, and wait for its reply.

• Agent API: create, update, message, activate, inspect, and archive persistent agents.

• Application requests: submit principal-addressed input and reconcile its owned responses behind a deployment gate.

• Inactive setup and named schedules: prepare agents, place them in Teams, and manage conditional configuration and schedules.

• REST files: upload, replace, inspect and download private file generations through a separately gated workflow.

• Service credentials: use Organization-owned machine identities with explicit resource grants and operation allowlists.

• Remote MCP: connect an MCP-capable client to the platform over HTTPS.

• Webhooks And Events: understand outbound task callbacks, inbound agent webhooks, provider webhooks, and app-trigger webhooks.

• Usage and audit metadata: scoped retained accounting and run reconciliation behind an independent deployment gate.

• Structured Data: request JSON Schema based outputs.

• Legacy Browser Tasks: use the older ad hoc browser-use task API when you need one-off task execution.

• Timeline and events: cursors, filters, pagination, and reply correlation.

• Pending actions: all six decision types and conflict handling.

• Errors and idempotency: recover and retry safely.

• MCP tool reference: generated schemas for every tool.

• SCIM: provision Organization users and groups.

• External Security Response API: request durable halts on explicitly scoped Organization agents and read enforcement evidence using dedicated service credentials; deployment-gated, with human recovery. For monitoring-only setup, use Configure External Security.

• Python security incident handler: a downloadable example with trusted asset mapping, persisted requests and evidence, safe retries, bounded polling, and human recovery.

• Halt while work is running: observe an admitted read, submit a halt, and distinguish worker cleanup from unresolved external outcomes.

• Limits and compatibility: supported boundaries and deployment settings.

Naming

End-user docs call the AI employee an agent. Developer docs use agent for the persistent API resource because that is the public API term.

Public integration surfaces

Prefer these surfaces for production integrations:

• /api/v1/agents/
• /api/v1/mcp/
• documented /api/v1/tasks/browser-use/ and /api/v1/agents/browser-use/... routes for legacy task use
• documented webhook URLs and payloads

The External Security Halt API adds documented /api/v1/security/interventions and /api/v1/security/targets/agent/{id}/control-state routes when the deployment and Organization gates are enabled. Its presence in the REST reference does not establish that a deployment is enabled or that its containment has been validated. See Availability and deployment gates.

Do not build production integrations against private console APIs unless the platform explicitly documents them as supported.