How trust-arbor/arbor works
Arbor is an Elixir/OTP platform for persistent, increasingly autonomous AI agents. A user’s chat enters the agent runtime, which maintains session context, invokes an LLM, and dispatches capability-scoped actions; memory and identity support continuity, while trust and security govern action authority. Separate product branches provide pipeline execution, consensus review, and monitoring/self-repair. The README documents these capabilities, but sampled source is partial, so some internal wiring is intentionally not shown.
Main entry points: executor.ex Action execution · agent.ex Agent runtime · llm_handler.ex LLM handler · provider_router.ex Provider routing
Components 21
Agent runtime
- Agent runtime
apps/arbor_agent/lib/arbor/agent.ex - Session manager
apps/arbor_agent/lib/arbor/agent/session_manager.ex - Action execution
apps/arbor_agent/lib/arbor/agent/executor.ex - Task orchestration
apps/arbor_agent/lib/arbor/agent/orchestration.ex
AI and memory
- LLM handler
apps/arbor_orchestrator/lib/arbor/orchestrator/handlers/llm_handler.ex - AI runtime
apps/arbor_ai/lib/arbor/ai/runtime.ex - Provider routing
apps/arbor_ai/lib/arbor/ai/runtime/provider_router.ex - Agent memory
- History queries
apps/arbor_historian/lib/arbor/historian/query_engine.ex
Trust and security
- Action catalog
apps/arbor_actions/lib/arbor_actions.ex - Trust policies
apps/arbor_actions/lib/arbor/actions/trust.ex - Security controls
apps/arbor_actions/lib/arbor/actions/security.ex
Workflows and governance
- Pipeline actions
apps/arbor_actions/lib/arbor/actions/pipeline.ex - Pipeline engine
apps/arbor_orchestrator/lib/arbor/orchestrator/engine.ex - Pipeline scheduler
apps/arbor_scheduler/lib/arbor/scheduler/workers/pipeline_runner.ex - Consensus council
apps/arbor_actions/lib/arbor/actions/council.ex
Operations
- Agent monitoring
apps/arbor_actions/lib/arbor/actions/monitor.ex - Self-remediation
apps/arbor_actions/lib/arbor/actions/remediation.ex
Other components
- User
- LLM provider
- Authoritative event log
Connections 17
- User → Agent runtime: sends request
- Agent runtime → Session manager: manages session
- Agent runtime → Task orchestration: coordinates tasks
- Task orchestration → LLM handler: runs LLM step
- LLM handler → AI runtime: requests completion
- AI runtime → Provider routing: routes request
- Provider routing → LLM provider: calls provider
- LLM provider → Provider routing: returns response
- Session manager → Agent memory: maintains context
- History queries → Authoritative event log: reads history
- Action execution → Action catalog: dispatches action
- Action execution → Trust policies: checks authority
- Action execution → Security controls: enforces controls
- Pipeline actions → Pipeline engine: runs pipeline
- Pipeline scheduler → Pipeline engine: executes attested graph
- Consensus council → LLM handler: requests perspectives
- Agent monitoring → Self-remediation: proposes fixes