Containerized AI agents

This repository provides single-file shell scripts that launch AI agents inside disposable Docker containers.

What this gives you

• No separate Dockerfile or docker-compose.yml is needed.
• The container image is defined directly inside each script.
• The current working folder is mounted to /workspace inside the container.
• Agent host configs are mounted into the container so settings like auth, model selection, and local state stay in sync.
• Ollama models can stay on the host machine while the agent runs in Docker.
• The container is removed automatically after the agent exits because it runs with --rm.

Available scripts

• opencode-agent.sh launches the OpenCode agent.
• pi-agent.sh launches the Pi coding agent.

How it works

Each script is self-contained:

1. It captures the current directory as the workspace.
2. It builds a minimal Docker image inline from a heredoc Dockerfile.
3. It mounts the workspace at /workspace.
4. It mounts the relevant host config directories so the agent keeps using your local settings.
5. It exposes the host machine to the container as host.docker.internal and sets OLLAMA_HOST so the agent can reach a host Ollama server.
6. It starts the agent in an interactive container that is deleted on exit.

Usage

Run the script you want from the repository root or from any project directory you want the agent to work on. All scripts support a -r/--rebuild flag to force rebuilding the container image before launching.

Examples:

opencode-agent.sh
pi-agent.sh

# Force rebuild of the container image
opencode-agent.sh -r
pi-agent.sh --rebuild

Because the current folder is mounted into the container, run the script from the project you want the agent to inspect or modify.

Using Ollama on the host machine

Both scripts are set up to use an Ollama server running on the host machine.

• By default, the container connects to http://host.docker.internal:11434.
• The scripts pass OLLAMA_HOST into the container automatically.
• If your Ollama server listens on a different port, set OLLAMA_PORT before launching the agent.

Examples:

# Default Ollama port
opencode-agent.sh

# Custom Ollama port on the host
OLLAMA_PORT=11500 pi-agent.sh

Make sure Ollama is running on the host before starting the containerized agent.

Notes

• The scripts expect Docker to be available locally.
• If you want to use local Ollama models, Ollama must also be running on the host machine.
• The first run may take longer because the image is built and dependencies are installed inside the container.
• Use the -r or --rebuild flag to force rebuilding the container image (useful after script updates or when you want to pick up upstream image or agent changes).
• Any settings stored in the mounted config directories will persist between runs even though the container itself is ephemeral.