System Settings

Available at Admin → Settings. Changes here affect all users on the server.

---

Organisation

Organisation name — display name shown in the UI and in emails.

Workspace ID — your subdomain ({slug}.bodhai.app). Cannot be changed after setup without re-provisioning your Cloudflare tunnel.

---

Language Model

Model — the Qwen model currently loaded. Changing this requires a server restart and model download (if the new model is not already cached).

LLM base URL — the endpoint BodhAI uses for inference. Default: http://localhost:8080/v1 (mlx-lm.server on Mac, llama-server on Linux). Advanced users can point this at a custom OpenAI-compatible endpoint.

Max concurrent requests — how many simultaneous LLM calls the server allows. Increase on high-RAM machines; decrease if you see out-of-memory errors under load.

---

Retrieval

Retrieval chunk count — number of chunks retrieved per query before re-ranking. Default: 20. Higher values improve recall but increase latency.

Reranker — the cross-encoder model used to re-rank retrieved chunks. Default: BAAI/bge-reranker-base (~278 MB, runs locally).

---

Conversation history

Store chat history — when enabled, conversations (including citations, generated files, and reasoning) are stored in the local SQLite database so users can browse and resume them — see Chat History. Enabled by default; turn it off to stop persisting conversations org-wide.

Retention period — conversations are deleted automatically after this many days. Default: 30.

---

Danger zone

Reset knowledge base — deletes all indexed documents and vector embeddings. Original uploaded files are preserved. Use this to start fresh without reinstalling.

Factory reset — removes all data including users, settings, and uploaded files. Requires admin password confirmation.