How xtremislv/myai works
AI Platform is a self-hosted AI service for authenticated chat and governed department agents, backed by a local vLLM endpoint. A chat request is authenticated, handled as private chat or routed through the Request Manager, then executed by an agent graph; critical tool actions can pause for approval. Other product branches provide policy-managed object storage, encrypted secrets and user chat keys, department configuration, and operational telemetry. Edges reflect README statements and source-index references; unsampled implementation details are not inferred.
Main entry points: router.py Request Manager · routes_chat.py Chat route · engine.py ACL engine · main.py FastAPI app
Components 24
API and identity
- FastAPI app
app/main.py - Auth routes
app/api/routes_auth.py - Chat route
app/api/routes_chat.py - Auth dependencies
app/api/deps.py - JWT security
app/auth/security.py
Request governance
- Request Manager
app/request_manager/router.py - Department classifier
app/request_manager/classifier.py - Approval store
app/request_manager/approvals.py - Department packs
app/departments/packs.py
Agent execution
- Graph execution
app/orchestrator/execution.py - Agent graphs
app/orchestrator/graph.py - Agent nodes
app/orchestrator/agents.py - Agent tools
app/orchestrator/tools.py - LLM client
app/llm/client.py
Data protection
- ACL engine
app/acl/engine.py - Object storage
app/storage/objects.py - Encrypted vault
app/vault/aesgcm_vault.py - User chat keys
app/vault/user_keys.py - Postgres data
app/db/session.py
Operations
- Telemetry
app/observability.py
Other components
- User
- vLLM service
- MinIO
- OTel and metrics
Connections 28
- User → FastAPI app: sends requests
README.md - FastAPI app → Auth routes: registers routes
app/main.py - FastAPI app → Chat route: registers routes
app/main.py - User → Auth routes: logs in
README.md - Chat route → Auth dependencies: resolves identity
app/api/routes_chat.py - Auth dependencies → JWT security: decodes token
app/api/deps.py - Auth routes → JWT security: uses token security
app/api/routes_auth.py - Chat route → Request Manager: routes work requests
app/api/routes_chat.py - Request Manager → Department classifier: classifies requests
app/request_manager/router.py - Request Manager → Department packs: checks department
app/request_manager/router.py - Request Manager → Approval store: checks pending approvals
app/request_manager/router.py - Request Manager → Graph execution: runs graph
app/request_manager/router.py - Graph execution → Agent graphs: selects graph
app/orchestrator/execution.py - Graph execution → Agent nodes: runs agents
app/orchestrator/execution.py - Agent nodes → Agent tools: uses tools
app/orchestrator/agents.py - Agent nodes → LLM client: requests completions
app/orchestrator/agents.py - LLM client → vLLM service: calls inference
README.md - Agent tools → ACL engine: checks access
app/orchestrator/tools.py - Agent tools → Object storage: uses object storage
app/orchestrator/tools.py - Object storage → ACL engine: enforces policy
app/storage/objects.py - Object storage → MinIO: reads and writes
app/storage/objects.py - ACL engine → Postgres data: reads policies
app/acl/engine.py - Encrypted vault → Postgres data: stores encrypted secrets
app/vault/aesgcm_vault.py - Chat route → User chat keys: uses chat keys
app/api/routes_chat.py - User chat keys → Postgres data: wraps user keys
app/vault/user_keys.py - LLM client → Telemetry: records metrics
app/llm/client.py - ACL engine → Telemetry: records decisions
app/acl/engine.py - Telemetry → OTel and metrics: exports telemetry
app/observability.py