How archestra-ai/archestra works
Archestra is an enterprise AI platform serving end users, agents, and developer clients through chat, LLM, and MCP entry points. A typical interaction is initiated in a frontend or client, handled by a backend route or proxy, sent to a model or tool, and governed by identity, permissions, and guardrails. The README also describes agent triggers and delegation, knowledge retrieval, mini apps, MCP/A2A gateways, and observability. Those documented capabilities are represented, but their detailed wiring is not established by the sampled source excerpts.
Main entry points: agent-run-stream.ts Agent Runtime · llm-proxy-handler.ts LLM Proxy · index.ts Platform Database · query-client-provider.tsx Web Frontend
Components 25
User Interfaces
- Web Frontend
platform/frontend/src/app/_parts/query-client-provider.tsx - Chat & Apps
platform/backend/src/routes/app/app.routes.ts - LLM OAuth UI
platform/frontend/src/components/llm-oauth-client-dialogs.tsx
Gateway Runtime
- OAuth Client Routes
platform/backend/src/routes/llm-oauth-clients.ts - LLM Proxy
platform/backend/src/routes/proxy/llm-proxy-handler.ts - Proxy Configuration
platform/backend/src/routes/llm-proxy/llm-proxy.routes.ts - Provider Client
platform/backend/src/clients/llm-client.ts
Agents & Tools
- Agent Runtime
platform/backend/src/agents/agent-run-stream.ts - MCP Gateway
platform/backend/src/archestra-mcp-server/index.ts - A2A Gateway
platform/backend/src/agents/a2a/a2a-manager.ts - Knowledge Base
platform/backend/src/database/schemas/knowledge-base.ts - Mini Apps
platform/backend/src/routes/app/app.routes.ts - Sandbox Execution
platform/archestra-rs/sandbox-core/src/lib.rs - MCP Orchestration
platform/backend/src/archestra-mcp-server/mcp-gateways.ts - Private MCP Registry
platform/backend/src/archestra-mcp-server/mcp-servers.ts
Security & Policy
- Guardrails
platform/backend/src/routes/openappa-coverage/openappa-coverage.routes.ts - Identity & Access
platform/backend/src/auth/index.ts - Environments & Limits
platform/backend/src/database/schemas/environment.ts
Platform State
- OAuth Client Model
platform/backend/src/models/llm-oauth-client.ts - Observability
platform/backend/src/database/schemas/interaction.ts - Platform Database
platform/backend/src/database/index.ts
Other components
- Users
- Developer Clients
- LLM Providers
- MCP Services
Connections 25
- Users → Web Frontend: use
- Developer Clients → LLM Proxy: request
- Web Frontend → Chat & Apps: invoke
- Web Frontend → LLM OAuth UI: manage clients
- LLM OAuth UI → OAuth Client Routes: configure
- OAuth Client Routes → OAuth Client Model: manage records
- Chat & Apps → Agent Runtime: start runs
- LLM Proxy → Provider Client: dispatch
- Provider Client → LLM Providers: call model
- LLM Proxy → Guardrails: apply policy
- LLM Proxy → Observability: record interactions
- Proxy Configuration → Platform Database: read/write config
- OAuth Client Model → Platform Database: read/write clients
- OAuth Client Routes → Identity & Access: authorize
- Agent Runtime → LLM Proxy: request inference
- Agent Runtime → MCP Gateway: invoke tools
- MCP Gateway → MCP Services: dispatch tools
- Agent Runtime → Knowledge Base: retrieve context
- Agent Runtime → Sandbox Execution: execute code
- MCP Orchestration → Private MCP Registry: manage servers
- A2A Gateway → Agent Runtime: trigger agents
- Agent Runtime → Observability: emit telemetry
- Guardrails → Platform Database: read policy state
- Identity & Access → Platform Database: read identity state
- Environments & Limits → Platform Database: persist limits