pygent 0.1.15

Pygent is a minimalist coding assistant that runs commands in a Docker container when available and falls back to local execution. See https://marianochaves.github.io/pygent for documentation and https://github.com/marianochaves/pygent for the source code.

Pygent

Pygent is a coding assistant that executes each request inside an isolated Docker container whenever possible. If Docker is unavailable (for instance on some Windows setups) the commands are executed locally instead. Full documentation is available in the docs/ directory and at marianochaves.github.io/pygent.

Features

• Runs commands in ephemeral containers (default image python:3.12-slim).
• Integrates with OpenAI-compatible models to orchestrate each step.
• Persists the conversation history during the session.
• Provides a small Python API for use in other projects.
• Optional web interface via pygent-ui.
• Register your own tools and customise the system prompt.

Installation

Installing from source is recommended:

pip install -e .

Python ≥ 3.9 is required. The package now bundles the openai client for model access. To run commands in Docker containers also install pygent[docker].

Configuration

Behaviour can be adjusted via environment variables (see docs/configuration.md for a complete list):

• OPENAI_API_KEY – key used to access the OpenAI API. Set this to your API key or a key from any compatible provider.
• OPENAI_BASE_URL – base URL for OpenAI-compatible APIs (defaults to https://api.openai.com/v1).
• PYGENT_MODEL – model name used for requests (default gpt-4.1-mini).
• PYGENT_IMAGE – Docker image to create the container (default python:3.12-slim).
• PYGENT_USE_DOCKER – set to 0 to disable Docker and run locally.
• PYGENT_MAX_TASKS – maximum number of concurrent delegated tasks (default 3).

CLI usage

After installing run:

pygent

Use --docker to run commands inside a container (requires pygent[docker]). Use --no-docker or set PYGENT_USE_DOCKER=0 to force local execution. Pass --config path/to/pygent.toml to load settings from a file.

Type messages normally; use /exit to end the session. Each command is executed in the container and the result shown in the terminal. Interactive programs that expect input (e.g. running python without a script) are not supported and will exit immediately. For a minimal web interface run pygent-ui instead (requires pygent[ui]).

API usage

You can also interact directly with the Python code:

from pygent import Agent

ag = Agent()
ag.step("echo 'Hello World'")
# ... more steps
ag.runtime.cleanup()

See the examples folder for more complete scripts. Models can be swapped by passing an object implementing the Model interface when creating the Agent. The default uses an OpenAI-compatible API, but custom models are easy to plug in.

Using OpenAI and other providers

Set your OpenAI key:

export OPENAI_API_KEY="sk-..."

To use a different provider, set OPENAI_BASE_URL to the provider endpoint and keep OPENAI_API_KEY pointing to the correct key:

export OPENAI_BASE_URL="https://openrouter.ai/api/v1"
export OPENAI_API_KEY="your-provider-key"

Development

1. Install the test dependencies:

pip install -e .[test]

2. Run the test suite:

pytest

Use mkdocs serve to build the documentation locally.

License

This project is released under the MIT license. See the LICENSE file for details.