# Agent AFK > Agent AFK is an open-source coding agent designed for when you're not watching. Start a task, walk away, and come back to a verified receipt of every step, every decision, and what's waiting on you. Permission boundaries, terminal states, trace history, verification workflows, and Telegram supervision mean you know exactly what happened before anything ships. Works with any model provider including local models. Apache-2.0. BYO model. ## What is an AFK Agent? An AFK agent is an AI coding agent designed to keep working after you walk away from the keyboard. "AFK" stands for "Away From Keyboard" — borrowed from gaming, applied to AI agents. The agent runs a task end to end, verifies its own output, and reaches a clear terminal state (Done, Blocked, Asking, or Interrupted), then notifies you asynchronously. Most coding agents can run without you present; an AFK agent is designed for it — with terminal states, self-verification, reversibility gates, and asynchronous communication built into the architecture. Agent AFK is the canonical open-source implementation of this pattern. Learn more: https://agentafk.com/what-is-an-afk-agent Install as the `afk` command on your PATH: ``` npm install -g agent-afk ``` Requires Node ≥20. Agent AFK attaches to your work through three surfaces — an interactive CLI REPL, an unattended background daemon, and a Telegram bridge for mobile supervision — all converging on an isolated `~/.afk/` root (config, plugins, plans, skills, session traces, and memory). Nothing leaks into your global config. Most coding agents work linearly by default — one hypothesis, one file, one pass — then stop at a vague "done," and when they do parallelize the work drifts without structure. Agent AFK changes the shape of the run: skills route work into parallel hypotheses, isolated worktrees, and verification loops by default, then govern it with traces, terminal states, and reversibility rules. The operating model is "delegate the work, keep the judgment": move fast on reversible work (isolated worktrees, scratch edits) and slow down on risky work — commits, pushes, deletes, and external calls are gated or verified. Every run resolves to one terminal state (Done, Blocked, Asking, or Interrupted), and plans, traces, and session state persist under `~/.afk/` so you can inspect, resume, or roll back. ## Capabilities - Subagent orchestration: Parallel and nested subagents by default — permission bubbling and transitive abort; one interrupt stops the whole tree. - Prompt & skill routing: One command dispatches a full parallel pipeline — each slash skill is a multi-stage workflow, not a single prompt you must over-specify. - Local trace history: Sessions, subagents, and tool calls under `~/.afk/`. Re-read, resume, or roll back. - Tool visibility: Inline tool calls with arguments, output, and timing — the path, not just the answer. - Verification workflows: Adversarial re-derivation of subagent claims before they hit your diff. - Remote supervision: Telegram bridge — stream output, send follow-ups, interrupt from your phone. ## Slash skills Each skill fans out subagents, gathers results, and re-checks before returning. Eleven ship in the box; `/forge` generates more. - /mint: End-to-end feature pipeline — spec → research → plan → parallelize → build → verify → heal → ship. - /diagnose: Parallel hypothesis generation and validation for bugs and failing tests. - /shadow-verify: Adversarial re-derivation of sub-agent claims before you act on them. - /forge: Generate new skills autonomously, gated by L1 capability evals. - /parallelize: Transform a linear plan into dependency-aware parallel waves. - /review: Parallel dimension agents across a diff or PR — security, correctness, api-compat, tests, perf. - /ship: Release pipeline for finished local work — pre-flight, tests, commit, push, PR. - /spec: Turn a loose idea into a structured, actionable spec ready for implementation. - /research: Two sub-agents in parallel — web context plus local repo context — merged into one brief. - /devils-advocate: Three parallel critics invent alternative approaches; a synthesis step ranks all four. - /gather: Parallel context-gathering for a code area you're about to edit. ## Architecture Three attach surfaces converge on one isolated root: - CLI REPL: Interactive TUI — streaming chat, plan mode, slash skills, background tasks. - Daemon: Background loop — runs unattended with bypass-permissions and plan persistence. - Telegram bridge: Mobile attach — send messages, tail output, interrupt from anywhere. - ~/.afk/: The hub — config, plugins, plans, skills, session traces, and memory. ## Pricing Free and open source (Apache-2.0) — the whole runtime, no feature held back. Bring your own model (Claude / API / provider access; tokens stay with your provider). Paid plans are ongoing monthly support from the maintainer (setup, tuning, custom skills, priority access) — not feature gates. Enterprise governance (SSO, audit logs, RBAC, policy gates) is in design. ## Links - [GitHub repository](https://github.com/griffinwork40/agent-afk): Source, issues, and documentation (Apache-2.0). - [npm package](https://www.npmjs.com/package/agent-afk): Install with `npm install -g agent-afk` (requires Node ≥20). - [What is an AFK Agent?](https://agentafk.com/what-is-an-afk-agent): Canonical definition and explainer. - [Homepage](https://agentafk.com/): Product overview and feature walkthrough. - [Changelog](https://agentafk.com/changelog): All notable changes to Agent AFK, newest first. - [Settings generator](https://agentafk.com/settings): Build a `~/.afk/` config without reading the docs — runs entirely in your browser. ## Optional - [Contact](mailto:griffin@graisol.com?subject=Agent%20AFK&body=Hi%20%E2%80%94%20): Setup help, workflow audits, and team / enterprise inquiries.