diff --git a/OLLAMA-STACK.md b/OLLAMA-STACK.md
new file mode 100644
index 0000000..4774a04
--- /dev/null
+++ b/OLLAMA-STACK.md
@@ -0,0 +1,109 @@
+# 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:
+
+```bash
+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:
+
+```bash
+curl http://localhost:11434/api/tags                                     # bash/zsh
+Invoke-RestMethod http://localhost:11434/api/tags                        # PowerShell
+```
+
+## Connect Cursor
+
+Two independent integrations. Enable either or both.
+
+### Notes via MCP
+
+Add to `~/.cursor/mcp.json`:
+
+```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`.
+
+```bash
+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.
+
+## Links
+
+- NoteDiscovery: https://github.com/gamosoft/NoteDiscovery
+- MCP tools reference: [`documentation/MCP.md`](documentation/MCP.md)
+- Ollama library: https://ollama.com/library
diff --git a/README.md b/README.md
index 1d1a3b8..e9b54f4 100644
--- a/README.md
+++ b/README.md
@@ -111,6 +111,8 @@ NoteDiscovery includes a built-in **Model Context Protocol (MCP)** server, letti
 ```
 
 > 💡 **See [MCP.md](documentation/MCP.md)** for complete setup instructions and all available tools.
+>
+> 🧪 **Want a fully local setup with a bundled LLM?** [OLLAMA-STACK.md](OLLAMA-STACK.md) spins up NoteDiscovery + Ollama + Open WebUI with one command.
 
 ## 📺 Watch the tour
 
@@ -154,6 +156,7 @@ Two docker-compose files are provided:
 |------|----------|
 | `docker-compose.ghcr.yml` | **Recommended** - Uses pre-built image from GitHub Container Registry |
 | `docker-compose.yml` | For development - Builds from local source |
+| `docker-compose.ollama-stack.yml` | Bundled local AI stack (NoteDiscovery + Ollama + Open WebUI) — see [OLLAMA-STACK.md](OLLAMA-STACK.md) |
 
 **Option 1: Pre-built image (fastest)**
 
diff --git a/docker-compose.ollama-stack.yml b/docker-compose.ollama-stack.yml
new file mode 100644
index 0000000..0fc1d4e
--- /dev/null
+++ b/docker-compose.ollama-stack.yml
@@ -0,0 +1,90 @@
+services:
+  notediscovery:
+    image: ghcr.io/gamosoft/notediscovery:latest
+    container_name: notediscovery
+    restart: unless-stopped
+    ports:
+      - "8000:8000"
+    environment:
+      - TZ=UTC
+    volumes:
+      - ./data:/app/data
+      # Uncomment to customize (files/folders must exist first):
+      # - ./config.yaml:/app/config.yaml
+      # - ./plugins:/app/plugins
+      # - ./themes:/app/themes
+    healthcheck:
+      test: ["CMD", "python", "-c", "import urllib.request; urllib.request.urlopen('http://localhost:8000/health')"]
+      interval: 30s
+      timeout: 5s
+      retries: 3
+      start_period: 10s
+
+  ollama:
+    image: ollama/ollama:latest
+    container_name: ollama
+    restart: unless-stopped
+    ports:
+      - "11434:11434"
+    environment:
+      - OLLAMA_HOST=0.0.0.0:11434
+      - OLLAMA_KEEP_ALIVE=30m
+    volumes:
+      - ollama-models:/root/.ollama
+    healthcheck:
+      test: ["CMD", "ollama", "--version"]
+      interval: 30s
+      timeout: 5s
+      retries: 3
+      start_period: 15s
+
+  # ChatGPT-style browser UI for Ollama.
+  # Available at http://localhost:3000 after startup.
+  open-webui:
+    image: ghcr.io/open-webui/open-webui:main
+    container_name: open-webui
+    restart: unless-stopped
+    ports:
+      - "3000:8080"
+    environment:
+      - OLLAMA_BASE_URL=http://ollama:11434
+      # Skip account creation on first launch (single-user local mode).
+      # Set to "true" for the login flow if you're exposing this beyond localhost.
+      - WEBUI_AUTH=false
+      # We only use Ollama, so disable the external OpenAI integration.
+      - ENABLE_OPENAI_API=false
+    volumes:
+      - open-webui-data:/app/backend/data
+    depends_on:
+      ollama:
+        condition: service_healthy
+
+  # One-shot service: pulls the model on first startup, then exits.
+  # No-op on subsequent runs if the model is already present.
+  # To change the model, edit MODEL below (any tag from https://ollama.com/library):
+  #   qwen2.5:0.5b        ~400 MB   fastest, minimal capability
+  #   qwen2.5:1.5b        ~1.0 GB   balanced (default)
+  #   qwen2.5-coder:1.5b  ~1.0 GB   code-focused
+  #   qwen2.5-coder:3b    ~2.0 GB   better code, still small
+  #   llama3.2:3b         ~2.0 GB   general-purpose
+  ollama-init:
+    image: ollama/ollama:latest
+    container_name: ollama-init
+    depends_on:
+      ollama:
+        condition: service_healthy
+    environment:
+      - OLLAMA_HOST=ollama:11434
+      - MODEL=qwen2.5:1.5b
+    entrypoint:
+      - /bin/sh
+      - -c
+      - |
+        echo "Pulling model: $$MODEL"
+        ollama pull "$$MODEL"
+        echo "Model ready."
+    restart: "no"
+
+volumes:
+  ollama-models:
+  open-webui-data:
diff --git a/documentation/MCP.md b/documentation/MCP.md
index 654c205..885cbc7 100644
--- a/documentation/MCP.md
+++ b/documentation/MCP.md
@@ -459,3 +459,7 @@ NoteDiscovery/
 │   └── tools.py         # Tool definitions
 └── ...
 ```
+
+## Try it with a local LLM
+
+For a batteries-included setup that runs NoteDiscovery alongside [Ollama](https://ollama.com) (local LLM runtime) and [Open WebUI](https://openwebui.com) (browser chat UI), see **[OLLAMA-STACK.md](../OLLAMA-STACK.md)** in the repo root. One `docker compose` command spins everything up, and you can point Cursor at both your notes (via this MCP server) and your local model.