4.3 KiB

Raw Blame History

NoteDiscovery + Ollama + Open WebUI

A one-command local AI stack. Runs three services, downloads a small model, no cloud, no API keys.

Service	URL	What it is
NoteDiscovery	http://localhost:8000	Markdown notes app
Open WebUI	http://localhost:3000	ChatGPT-style browser UI
Ollama	http://localhost:11434	Local LLM runtime (OpenAI-compatible at `/v1`)

Default model: qwen2.5:1.5b (~1 GB). Change it in docker-compose.ollama-stack.yml under ollama-init.

Prerequisites

Docker Desktop (Windows/macOS) or Docker Engine + Compose v2 (Linux)
~5 GB free disk, ~2 GB free RAM

Start

From the repo root:

docker compose -f docker-compose.ollama-stack.yml up -d
docker compose -f docker-compose.ollama-stack.yml logs -f ollama-init   # wait for "Model ready."

First run pulls images + the model (5–10 min). After that it's seconds.

Then open:

http://localhost:8000 — start taking notes (saved to ./data/)
http://localhost:3000 — pick qwen2.5:1.5b and chat

Quick sanity-check that Ollama is up:

# List installed models
curl http://localhost:11434/api/tags                                     # bash/zsh
Invoke-RestMethod http://localhost:11434/api/tags                        # PowerShell

# Actually poke the model
curl http://localhost:11434/api/generate -d '{"model":"qwen2.5:1.5b","prompt":"Finish the joke in one short sentence: Why do programmers prefer dark mode? Because...","stream":false}'                    # bash/zsh
Invoke-RestMethod http://localhost:11434/api/generate -Method Post -Body '{"model":"qwen2.5:1.5b","prompt":"Finish the joke in one short sentence: Why do programmers prefer dark mode? Because...","stream":false}'   # PowerShell

Connect Cursor

Two independent integrations. Enable either or both.

Notes via MCP

Add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "notediscovery": {
      "command": "docker",
      "args": [
        "run", "--rm", "-i",
        "-e", "NOTEDISCOVERY_URL=http://host.docker.internal:8000",
        "ghcr.io/gamosoft/notediscovery:latest",
        "python", "-m", "mcp_server"
      ]
    }
  }
}

Restart Cursor. Ask things like "list my recent notes" or "create a note called scratch with today's date".

Linux: host.docker.internal doesn't resolve by default. Add "--add-host=host.docker.internal:host-gateway" to the args array.

Local model as Cursor's chat model

Cursor Settings → Models → add a custom OpenAI-compatible model:

Base URL: http://localhost:11434/v1
API Key: anything (e.g. ollama)
Model: qwen2.5:1.5b

Only affects the chat model picker — Cursor Tab and background jobs still use Cursor's cloud models. If Cursor rejects the local URL (older versions validate from the cloud), expose it via cloudflared tunnel --url http://localhost:11434 and use the tunnel URL instead.

Useful commands

All commands below assume COMPOSE_FILE=docker-compose.ollama-stack.yml is set; otherwise prepend -f docker-compose.ollama-stack.yml.

docker compose logs -f <service>          # tail logs
docker compose exec ollama ollama list    # list installed models
docker compose exec ollama ollama pull qwen2.5-coder:3b   # add a model
docker compose down                       # stop (keeps everything)
docker compose down -v                    # stop + wipe models & chat history (notes survive)
docker compose pull && docker compose up -d   # update

Any model tag from https://ollama.com/library works. Rough RAM rule: 1B ≈ 1 GB, 3B ≈ 2 GB, 7B ≈ 5 GB.

Data

./data/ — your notes (plain markdown, back this up)
ollama-models volume — models (re-downloadable)
open-webui-data volume — chat history (safe to wipe)

Troubleshooting

Open WebUI shows "no models" — model pull hasn't finished. Watch docker compose logs -f ollama-init.
Cursor MCP disconnected — check curl http://localhost:8000/health, then restart Cursor.
Slow generation — use a smaller model (qwen2.5:0.5b) or give Docker more CPU/RAM.
Port already in use — edit the ports: mapping in docker-compose.ollama-stack.yml, then update your MCP config / URLs.

4.3 KiB Raw Blame History Unescape Escape