Running an AI agent once is easy. Running one every hour, unsupervised, while tracking what it spent and what it did - that's the problem botctl is trying to solve.
botctl is an open-source process manager for autonomous AI agents, written in Go and built around Claude. You define a bot in a BOT.md file using YAML for configuration (schedule, max turns, session memory) and plain markdown for the actual prompt. The tool handles the loop: spawn Claude with your prompt and tools, process the results, save session state, sleep, repeat. A terminal dashboard and a browser-based interface let you watch logs, track cost-per-run, and redirect an active bot mid-execution without stopping it.
The design mirrors tools like PM2 or supervisord for Node.js - if you've used those to keep web servers alive, the mental model is the same, except your "process" is a Claude agent doing actual work. Skills (reusable tool modules) can be pulled in from GitHub repos, which keeps individual bot configs lean. Hot-reload means you can edit a prompt and have it take effect on the next loop without restarting anything.
It installs as a single binary on macOS, Linux, and Windows (AMD64 and ARM64). There's no hosted service - you run it yourself, on your own machine or a server, against your own Anthropic API key.
The practical use cases are narrow but real: scheduled content checks, automated monitoring agents, workflows that need to run on a timer without you babysitting a chat window. For developers already comfortable with the command line and Claude's API, botctl removes the boilerplate of building that scheduling infrastructure from scratch.