Roadmap
Alan is developed in phases, each building on the previous one.
| Phase | Goal | Status |
|---|---|---|
| Phase 1 | Working end-to-end pipeline | Current |
| Phase 2 | Improved retrieval quality and production-ready server | Planned |
| Phase 3 | Agentic capabilities and MCP integration | Planned |
| Phase 4 | Multi-graph support | Future |
| Phase 5 | MCP server for external AI agents | Future |
Current status — Phase 1
The end-to-end pipeline is operational:
- Intent routing:
lookup/rag/agent - Vector similarity search via TuringDB's built-in vector index
- Graph traversal via Cypher queries
- Agentic loop with 7 tools (vector search, graph query, 5 schema tools)
- Graph skeleton injected into the system prompt at startup
- Chat session persistence in Valkey
- Per-user request and token rate limiting
- Streamlit web UI
- TUI for service management
See Phase 1 for the full feature list and open items.
What's next — Phase 2
The main focus of Phase 2 is retrieval quality and production readiness:
- Richer context retrieval (neighbor expansion, publication enrichment)
- Replace Streamlit with a proper React frontend
- LLM provider switching per-request (no restart required)
- SSE streaming for intent router reasoning and per-step latencies
- PostgreSQL for user accounts and persistent chat history
See Phase 2 for the full plan.