From "Works on My Machine" to Works for the Whole Team.

Open Notebook is an open-source, self-hosted alternative to Google NotebookLM. It ingests documents (PDFs, URLs, YouTube transcripts, audio), indexes them with embeddings, and lets you chat with your content — plus generate multi-speaker AI podcasts. It supports 18+ AI providers instead of just Gemini, has no 50-source limit, and can run entirely offline.

The question isn't whether it's good. The question is whether you should deploy it.

The 2-Minute Test

mkdir open-notebook && cd open-notebook
curl -O https://raw.githubusercontent.com/lfnovo/open-notebook/main/docker-compose.yml
# Change the encryption key to your own secret string
sed -i 's/change-me-to-a-secret-string/a-random-string-you-pick-now/' docker-compose.yml
docker compose up -d

Open http://localhost:8502. On first run, Docker downloads the images (~30 seconds to 2 minutes depending on your connection). Once the interface appears, you've deployed Open Notebook.

The Honest Comparison

vs AnythingLLM

AnythingLLM (also open source) is similar — but Open Notebook's podcast/audio overview feature is the differentiator. If you don't need podcast generation, AnythingLLM has a more polished UI and easier setup. If you do need podcasts, Open Notebook is the clear winner.

vs PrivateGPT

PrivateGPT focuses on document Q&A with guaranteed local processing. It's lighter but has no podcast feature, no multi-model routing, and a more basic UI. Open Notebook is the full research workspace; PrivateGPT is the focused document chatbot.

Who Should Self-Host

Do it if:

• You handle sensitive documents (legal, medical, proprietary code) and can't upload them to Google
• You want API access to automate document processing in your workflow
• You've hit NotebookLM's 50-source cap and still have 100 more papers to ingest
• You need podcasts with more than 2 speakers or custom voice profiles
• You want to use Claude or GPT-4o for analysis quality that Gemini can't match
• You're building a product and need a white-label document AI backend

Skip it if:

• You use NotebookLM casually and have no compliance requirements
• You don't want to spend 1-2 hours on initial setup and ongoing maintenance
• You need production-quality citations (Open Notebook's citations are basic)
• You just need a chatbot for a few PDFs (use ChatGPT file upload or Claude Projects)

The $0 vs $30 vs $150 Setup

Note: Groq offers a generous free tier — most solo users won't hit $30/month under normal usage. The $30 figure is a conservative ceiling.

Chapter 0 gave you a 2-minute quickstart. This chapter explains every single line you just ran, so when something breaks at 2 AM, you know exactly where to look.

The Full docker-compose.yml — Line by Line

Here's the actual compose file from lfnovo/open-notebook, annotated:

services:
  # ── SERVICE 1: SurrealDB (Database) ──
  surrealdb:
    image: surrealdb/surrealdb:v2
    command: start --log info \
      --user ${SURREAL_USER:-root} \
      --pass ${SURREAL_PASSWORD:-root} \
      rocksdb:/mydata/mydatabase.db
    user: root  # Required for bind mounts on Linux
    ports:
      - "8000:8000"
    volumes:
      - ./surreal_data:/mydata     # All notebooks, notes, chat history
    environment:
      - SURREAL_EXPERIMENTAL_GRAPHQL=true
    restart: always
    pull_policy: always

  # ── SERVICE 2: Open Notebook App ──
  open_notebook:
    image: lfnovo/open_notebook:v1-latest
    ports:
      - "8502:8502"  # Web UI
      - "5055:5055"  # REST API
    volumes:
      - ./notebook_data:/app/data   # Uploaded files, processed docs
    environment:
      # REQUIRED: Change this to your own secret string
      - OPEN_NOTEBOOK_ENCRYPTION_KEY=change-me-to-a-secret-string
      # Database connection
      - SURREAL_URL=ws://surrealdb:8000/rpc
      - SURREAL_USER=${SURREAL_USER:-root}
      - SURREAL_PASSWORD=${SURREAL_PASSWORD:-root}
      - SURREAL_NAMESPACE=open_notebook
      - SURREAL_DATABASE=open_notebook
    depends_on:
      - surrealdb
    restart: always
    pull_policy: always

• No version: key — Docker Compose V2 doesn't need it.
• DB service starts first (depends_on), but the app retries internally on startup.
• user: root on SurrealDB is intentional — Linux bind mounts default to root ownership.
• No AI provider keys here. Configure them via the UI at Settings → API Keys. The old env-var approach is deprecated.

What Each Volume Actually Stores

You can delete ./notebook_data/ and re-upload documents without losing your research structure. You absolutely cannot delete ./surreal_data/.

The Networking Model

Inside Docker's default bridge network, services reach each other by container name:

open_notebook → surrealdb    via ws://surrealdb:8000/rpc    ← Docker DNS
open_notebook → ollama       via http://ollama:11434         ← Docker DNS
your browser  → open_notebook via http://localhost:8502      ← Host port mapping

Adding Ollama to the Stack

  ollama:
    image: ollama/ollama:latest
    container_name: open-notebook-ollama
    ports:
      - "11434:11434"
    volumes:
      - ./ollama_data:/root/.ollama
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]
    restart: always
    pull_policy: always

If you're on CPU-only or Apple Silicon, remove the entire deploy.resources block.

The First-Run Checklist

# 1. All services running?
docker compose ps

# 2. Web UI accessible?
curl -I http://localhost:8502

# 3. API responding?
curl http://localhost:5055/api/health

# 4. Database connected?
docker compose logs surrealdb | grep -i "started"

# 5. Ollama model loaded (if using Ollama)?
curl http://localhost:11434/api/tags

# 6. ENCRYPTION_KEY changed from default?
docker compose exec open_notebook env | grep ENCRYPTION_KEY

# 7. Disk space adequate?
df -h ./surreal_data ./notebook_data ./ollama_data

Five Common First-Run Errors (And Their Exact Fixes)

Error 1: Ports are not available

sudo lsof -i :8502
# Change host port: "8503:8502" → access on :8503

Error 2: PermissionError on SurrealDB

The SurrealDB container runs as root by default (user: root in compose), which is correct for Linux bind mounts. If you've changed this:

sudo chown -R 1000:1000 ./surreal_data

Error 3: Ollama embedding — Connection error, but chat works

This is the Esperanto library bug (Issue #655). In Settings → API Keys → Ollama, verify the Base URL is set correctly. If the embedding-specific URL field is available, set it explicitly. Workaround: use a cloud embedding provider (OpenAI text-embedding-3-small or Voyage AI) at < $0.01 per 1000 documents.

Error 4: CUDA out of memory

Error 5: WebSocket connection to SurrealDB failed

docker compose exec open_notebook sh -c "wget -qO- http://surrealdb:8000/health"
docker compose ps  # Verify both services are Up

SurrealDB: Quick Backup/Restore

# Backup
docker compose exec surrealdb surreal export \
  --conn http://localhost:8000 --user root --pass root \
  --ns open_notebook --db open_notebook \
  > backup-$(date +%Y%m%d).surql

# Restore
cat backup-20260625.surql | docker compose exec -T surrealdb surreal import \
  --conn http://localhost:8000 --user root --pass root \
  --ns open_notebook --db open_notebook

No. This is an independent community resource. Not affiliated with or endorsed by lfnovo or Google. It's built from real-world deployment experience — running Open Notebook in production for document research.

Email support@localnotebook.dev within 30 days for a full refund. No questions asked. We mean it — if the manual doesn't save you 20+ hours, you shouldn't pay for it.

Yes. You get all future updates at no extra cost. When Open Notebook ships a breaking change or new feature, the manual is updated within 2 weeks.

For personal use with cloud APIs, the free deployment guide is enough. If you're deploying for a team, running local models, or need it to work reliably day after day — that's when the 30+ error fixes, monitoring setup, and CI/CD templates pay for themselves.