How asadulm/agora works
AGORA is a self-hosted, document-grounded multi-agent debate platform. Users authenticate, configure agents and documents, start a debate, then inspect its staged reasoning, evidence, and final moderator verdict in a real-time workspace; follow-up cycles extend that discussion. The graph emphasizes the documented execution and RAG paths, plus authentication, presets, and live updates. Some backend routes and integrations are unsampled, so their connections are included only where the README or source index supports them.
Main entry points: round_manager.py Debate stages · main.py FastAPI application · ui Debate workspace · chat_engine.py Turn orchestration
Components 23
Web client
- Debate workspace
client/src/features/debate/ui - Debate state
client/src/features/debate/model/debate.store.ts - Authentication client
client/src/services/authService.ts - Agent presets
client/src/features/agent-presets
API and identity
- FastAPI application
server/app/main.py - Authentication routes
server/app/api/routes/auth.py - Debate routes
server/app/api/routes/debate.py - Document routes
server/app/api/routes/documents.py - Preset routes
server/app/api/routes/agent_presets.py - WebSocket routes
server/app/api/routes/ws.py
Debate execution
- Turn orchestration
server/app/services/chat_engine.py - Debate stages
server/app/services/debate_engine/round_manager.py - Follow-up cycles
server/app/services/followup_runner.py - Topic validation
server/app/services/topic_guard/topic_guard_service.py
Documents and retrieval
- Document ingestion
server/app/services/documents/ingestion_service.py - Evidence retrieval
server/app/services/retrieval/retrieval_service.py - Embedding service
server/app/services/embeddings/embedding_service.py
Models and integrations
- PostgreSQL and pgvector
- LLM provider router
server/app/services/llm - WebSocket manager
server/app/services/ws_manager.py - Document storage
server/app/services/storage - LLM providers
server/app/services/llm/providers
Other components
- User
Connections 29
- User → Debate workspace: uses
README.md - Debate workspace → Debate state: updates
client/src/features/debate/model/debate.store.ts - Debate workspace → Debate routes: starts debates
README.md - Debate workspace → Document routes: uploads documents
README.md - Debate workspace → WebSocket routes: subscribes live
README.md - Authentication client → Authentication routes: authenticates
README.md - Agent presets → Preset routes: loads presets
README.md - FastAPI application → Authentication routes: registers
server/app/main.py - FastAPI application → Debate routes: registers
server/app/main.py - FastAPI application → Document routes: registers
server/app/main.py - FastAPI application → Preset routes: registers
server/app/main.py - FastAPI application → WebSocket routes: registers
server/app/main.py - Debate routes → Turn orchestration: starts execution
README.md - Turn orchestration → Debate stages: runs stages
server/app/services/chat_engine.py - Turn orchestration → PostgreSQL and pgvector: loads and saves
README.md - Turn orchestration → WebSocket manager: publishes progress
README.md - Debate stages → Evidence retrieval: requests evidence
server/app/services/retrieval/retrieval_service.py - Debate stages → LLM provider router: generates responses
README.md - Debate stages → PostgreSQL and pgvector: persists stages
README.md - Debate stages → WebSocket manager: emits progress
README.md - Evidence retrieval → Embedding service: embeds query
server/app/services/retrieval/retrieval_service.py - Evidence retrieval → PostgreSQL and pgvector: searches chunks
server/app/services/retrieval/retrieval_service.py - Document routes → Document ingestion: processes uploads
server/app/services/documents/ingestion_service.py - Document ingestion → Embedding service: embeds chunks
server/app/services/documents/ingestion_service.py - Document ingestion → Document storage: stores originals
server/app/services/documents/ingestion_service.py - Document ingestion → PostgreSQL and pgvector: stores document data
server/app/services/documents/ingestion_service.py - LLM provider router → LLM providers: routes requests
server/app/services/llm/service.py - Follow-up cycles → Debate stages: executes follow-up
README.md - Debate routes → Topic validation: validates topics
README.md