Hani (goose in Estonian)

This project sets up Goose in a containerized environment, powered by the Model runner in Docker Desktop and connecting to the AI Tools Catalog by Docker.

If you have a Docker Desktop with the Model runner installed, you can run it locally:

docker run -it --rm -v /var/run/docker.sock:/var/run/docker.sock -p 7681:7681 olegselajev241/hani:local

Override the default model selection:

docker run -it --rm -v /var/run/docker.sock:/var/run/docker.sock -p 7681:7681 -e GOOSE_MODEL=ai/mistral-nemo:12B-Q4_K_M olegselajev241/hani

Prefer tools-aware models, Goose is an MCP client, so giving it access to tools multiplies the usefullness by a lot.

Compose based setup

If you have a Docker environment with GPU, you can run the compose based setup. It uses a Google Maps MCP server, so get your GOOGLE_MAPS_API_KEY, and put it into the .env file:

echo GOOGLE_MAPS_API_KEY="changeme" > .env

Then spin up the compose:

docker compose up

Cloud run based setup

You can also deploy it to Google Cloud Run. The project setup is a bit involved, you need a GPU quota, a VPC and the secret with the Maps API KEY, but you can check an example of the deployment descriptor in service.yaml.

Components of the Hani Docker Image

• Goose: Main application
• ttyd: Web terminal interface
• Docker CLI: For running socat container to access AI Tools catalog

Setup

1. Clone this repository:

   git clone https://github.com/shelajev/hani.git
   cd hani
   

2. Build all the container flavors:

   ./build.sh
   

3 Run the container locally:

docker run -it --rm -v /var/run/docker.sock:/var/run/docker.sock -p 7681:7681 olegselajev241/hani:local

4. Open Goose session in the browser: http://localhost:7681