codel

Why we included this project

Hand a long, multi-step job to Codel and a single agent carries it through. It can clone a repository, pull up the relevant docs in its built-in browser, edit files in the in-browser text editor, and run commands to check whether a fix actually works, picking the next action itself based on what it has done so far. Everything it touches runs inside a sandboxed Docker container, so file and system changes stay contained; that makes it a reasonable candidate to try before you trust any agent near a real environment. Because it is self-hosted and accepts OpenAI-compatible endpoints as well as locally hosted Ollama models, you decide where it runs and which model does the thinking. It also keeps every command and its output in a PostgreSQL database, so you can review exactly what it did.