A multi-agent knowledge operating system built on top of Open Notebook.
Powerful underneath. Simple on the surface.
This project is an independent extension of Open Notebook by Luis Novo, the open-source, privacy-focused alternative to Google's Notebook LM.
We build on the Open Notebook foundation — the three-tier architecture (Next.js → FastAPI → SurrealDB), multi-provider AI (18+ providers via Esperanto), multi-speaker podcasts, vector/text search, and the 14-locale UI — and add a multi-agent orchestration layer on top of it.
Upstream: https://github.com/lfnovo/open-notebook · Website: https://www.open-notebook.ai · License: MIT (
LICENSE, copyright © Luis Novo)
| Engine | What it does |
|---|---|
| Agent Engine | A supervisor (plan → execute → decide → finalize) orchestrates 10 specialized agents across a safe tool registry |
| Control Layer | Labels every claim verified / external / inferred / unverified; contradiction & hallucination checks |
| Research | gather → synthesize → fact-check produces source-cited reports with verification labels |
| Education | Turns any source into a study plan, explanation, quiz and flashcards |
| Persona | Re-interprets material from any expert perspective |
| Live Sync | Diffs external sources (Google Drive) into your knowledge base, deduped by content hash |
| Action + Approvals | External actions (email, Jira) execute only after human approval |
| Workflow + Scheduler | Run registered tool-step workflows on demand or hourly/daily/weekly |
| Multi-user foundation | user table + users API |
| Redesigned UI | Modern indigo design system, Home dashboard, non-technical navigation |
Read the full story in docs/agentic/:
flowchart LR
UI[Next.js 16<br/>:3000] --> API[FastAPI<br/>:5055]
API --> DB[(SurrealDB<br/>:8000)]
API -->|submit_command| Q[(Command queue)]
Q --> W[Background worker]
subgraph Engine layer
A[Agent supervisor] --> T[Tool registry]
R[Research] --> C[Control / verification]
E[Education]
P[Persona]
ACT[Action] --> AP[Approvals]
S[Live sync]
WF[Workflows + scheduler]
end
W --> A
W --> WF
T --> DB
Every long-running operation is a surreal-commands background job — the API
never blocks on podcast generation, embeddings, source processing or agent runs.
- Private & self-hosted — your data stays yours
- Multi-provider AI — OpenAI, Anthropic, Ollama, LM Studio and 15+ more
- Multi-agent orchestration — delegate a goal, the supervisor does the rest
- Verification layer — every claim gets an evidence label
- Multi-modal sources — PDFs, video, audio, web pages
- Multi-speaker podcasts — 1–4 speakers with custom profiles
- Full-text + vector search across everything
- Context-aware chat with citations
- Study material generation — plans, quizzes, flashcards
- Live sync & actions — Google Drive, email, Jira
- 14-locale UI
git clone https://github.com/koraysrn/agentic-open-notebook.git
cd agentic-open-notebook
docker compose upOpen http://localhost:3000. Configure your AI models in Settings → Models (or set API keys in the UI). For local models, start Ollama and run:
ollama pull qwen2.5 # chat model
ollama pull nomic-embed-text # embedding modelThen add both in Settings → Models (provider ollama). A helper script is
available for headless setup:
uv run python scripts/configure_ollama_models.pyThree tiers, started in order:
make database # SurrealDB
make api # FastAPI (migrations run automatically)
make worker-start # surreal-commands worker (podcasts/embeddings/agents)
make frontend # Next.js UIOr all at once: make start-all · status: make status · stop: make stop-all.
See Open Notebook's docs and docs/1-INSTALLATION/ for full installation and configuration guides.
uv run pytest tests/ # 761 backend tests
cd frontend
npm run test # 140 frontend tests
npm run lint
npm run buildCI runs the same suite on every push (see
.github/workflows/test.yml).
Python · FastAPI · Next.js · React · SurrealDB · LangChain · LangGraph · Esperanto
Distributed under the MIT License — see LICENSE.
This project is a derivative of Open Notebook © Luis Novo
(https://github.com/lfnovo/open-notebook), used under the MIT License. The
original copyright notice is preserved in LICENSE.