codalotl is a coding agent for Go codebases. Twice as fast and token efficent as standard agents. Deep integration with standard Go tooling and conventions.
codalotl is a Go-focused coding agent built to work in large (or small) Go codebases. It is primarily a TUI, but also comes with noninteractive CLI modes. It's like Codex/Claude Code/OpenCode (and can use the same models), but the harness itself just works better because it's purpose-built for Go.
Answer one: because I mostly write Go, and I primarily write codalotl for myself.
Answer two: there's a lot of things that can be done when you build an agent for just one language. To take a small example: codalotl automatically uses gofmt on all edits the LLM makes. The LLM never needs to think about this or make separate tool calls. I just works, 100% of the time.
Yes. codalotl is free to use, and there are no plans to add monetization of any kind. You are also free to fork the repository and adapt it for your own workflows.
Yes. The source is available at github.com/codalotl/codalotl, so you can inspect how it works, run it yourself, and build on top of it.
No. There are no codalotl accounts.
You just need API keys for your LLM provider, and codalotl calls the provider directly from your environment.
Right now, codalotl supports OpenAI models. We chose that starting point because OpenAI currently shows the best performance in goagentbench for the workflows codalotl is targeting.
Yes. codalotl servers do not see your code and do not proxy your LLM requests. Requests go directly from your environment to your configured model provider.
