So I built Kodama. It’s a self-hosted daemon written in Go that wraps these CLIs as subprocesses.

Key Features:

Asynchronous Execution: Queue tasks for multiple projects; Kodama runs them sequentially in the background.

Human-in-the-loop (Telegram): When the agent hits a KODAMA_QUESTION or gets stuck, it pings you via Telegram. You can reply directly to the bot to provide the answer.

Rate Limit Management: If the agent signals a rate limit, Kodama saves a checkpoint and automatically retries after a cooldown (default 5h).

Execution Runtimes: You can run project commands (build/test/lint) directly on your host or in an auto-scaffolded Docker environment.

Web UI: A simple dashboard to manage your backlog and watch the live output via WebSockets.

A note on Security: Kodama is designed as a local-first, trusted-network tool. There is no built-in authentication by design. Please run it on localhost, via a VPN (Tailscale/Wireguard), or behind a secure tunnel like Cloudflare Access.

The project is still a WIP, but it has already changed how I handle larger refactors or boilerplate-heavy features.

I’d love to get your feedback