A lightweight Docker stack that brings together Open‑WebUI, Searxng, n8n, PostgreSQL with the pgvector extension, Valkey, and Docling.
The stack is ready to run locally with a single docker compose up command.
git clone https://github.com/ikhalilov/LocalAIEnvironment.git
cd local-ai-env
cp .env.example .env # copy default configuration
docker compose up -d # start all services in the backgroundWait a few seconds for the containers to finish initializing. You can now access the UI at:
- Open WebUI – http://localhost:3000
- Docling – http://localhost:5001 (http://localhost:5001)
- Searxng – http://localhost:8080 (http://localhost:8080)
- n8n – http://localhost:5678 (http://localhost:5678)
The stack uses a single .env file. The repository ships a .env.example that contains sane defaults. Create a copy:
cp .env.example .env
| Variable | Description | Default |
|---|---|---|
| POSTGRES_USER | DB username | postgres |
| POSTGRES_PASSWORD | DB password | postgres |
| POSTGRES_DB | Database name | postgres |
| VALKEY_HOST | Valkey service name | valkey |
| VALKEY_PORT | Valkey port | 6379 |
| VALKEY_PASSWORD | Valkey password | password |
| SEARXNG_PORT | Searxng HTTP port | 8080 |
| SEARXNG_BIND_ADDRESS | Searxng bind address | 0.0.0.0 |
| DOCLING_SERVE_ARTIFACTS_PATH | Path to Docling models | /opt/app-root/src/.cache/docling/models |
| DOCLING_OCR_ENGINE | OCR engine to download | easyocr |
| CUDA_VISIBLE_DEVICES | GPU indices for Docling | 0,1 |
Other values are set directly in docker-compose.yml via the environment section. Modify the file if you need different ports or container names.
| Service | Purpose |
|---|---|
| Open WebUI | Chat UI that talks to LLMs via the Ollama endpoint (OLLAMA_BASE_URL). RAG is enabled and uses Searxng as the search backend. WebSocket support uses Valkey. |
| Searxng | Privacy‑focused search engine that powers the RAG component. Configuration lives in ./searxng_data. |
| n8n | Visual workflow automation tool that can trigger the Open WebUI or Docling services. |
| PostgreSQL + pgvector | Stores user data and embeddings. Image pgvector/pgvector:0.8.1-pg18-trixie includes the pgvector extension. |
| Valkey | Redis‑compatible store for websockets. |
| Docling | OCR/Document‑understanding server with GPU support. The command section downloads the necessary models on first start. |
- Add a new LLM – Update OLLAMA_BASE_URL or swap the container image that exposes the same API.
- Change the search engine – Edit RAG_WEB_SEARCH_ENGINE and SEARXNG_QUERY_URL in docker-compose.yml.
- Persist data – Docker volumes at the bottom of the compose file are retained across restarts.
- GPU usage – Ensure Docker Desktop has GPU support enabled and adjust CUDA_VISIBLE_DEVICES if you need a different GPU set.
- Open WebUI (https://github.com/open-webui/open-webui)
- Searxng (https://github.com/searxng/searxng)
- n8n (https://github.com/n8n-io/n8n)
- Docling (https://github.com/docling-project/docling-serve)
- PostgreSQL + pgvector (https://github.com/pgvector/pgvector)
- Valkey (https://github.com/valkey-io/valkey)
Feel free to open issues or pull requests. The repo is intentionally minimal, so improvements that make the stack more robust or easier to configure are welcome.
This project is provided as‑is under the MIT License. See the LICENSE (LICENSE) file for details.