
Agent AFK is an open-source, local-first runtime built for running coding agents unattended. Start a coding task, walk away, and come back to verified results you can actually trust—reviewed through activity traces, terminal states, and receipts of everything the agent did while you were gone.
Instead of babysitting every step, you define the guardrails once: permission boundaries, reversibility gates, isolated worktrees, parallel subagents, adversarial verification, and persistent session memory all work together to keep autonomous runs safe and under control. When you want a closer look, Telegram supervision lets you check in on your agents from anywhere.
Agent AFK works with the tools you already use, including Claude, Cursor, Claude Code, Codex, custom agents, and other model providers. Everything stays on your machine—the runtime and your data never leave it, so you get hands-off productivity without giving up privacy or control.

Subscribe to get the latest Agent AFK and global AI updates.
Comments
Loading comments…