Codel

Codel is a fully autonomous AI agent that performs complicated tasks and projects using a terminal, a browser, and a text editor. It is self-hosted and aimed at developers who want an agent that can work through multi-step projects on its own rather than responding to single prompts. Everything runs inside a sandboxed Docker environment for security.

Codel automatically detects the next step in a task and performs it without manual direction, picking a Docker image automatically based on the task at hand. A user runs a pre-built Docker image with an OpenAI or Ollama model configured through environment variables, then opens localhost:3000 in a browser to start using Codel. All history, commands, and outputs are saved in a PostgreSQL database.

Codel includes a built-in browser that fetches the latest information from the web, such as tutorials and documentation, when a task needs it, and a built-in text editor for viewing all the files it has modified directly in the browser. It supports OpenAI's GPT-4 models as well as locally hosted Ollama models such as Llama2, selected through environment variables when the Docker image is run. The interface is described by the project as a modern UI.

Codel is released under the AGPL-3.0 license and distributed through a Github Container Registry Docker image rather than a hosted service. Users must supply an environment variable for whichever language model they want to use, such as an OpenAI API key for GPT-4 or a locally hosted Ollama model, and can swap the OpenAI server URL to point at a compatible alternative endpoint. Support and discussion happen through the project's Discord server.

Use cases

• Completing multi-step coding projects without manual prompting
• Looking up documentation or tutorials mid-task