agentafk
Open-source local-first coding agent you can leave unattended with confidence.Start the task, walk away, get a verification receipt.
Highlights
What you can do with it
- Start a feature task, attend a meeting, and return to a completion notice
- Run a large overnight refactor and audit the local history next morning
- Manage coding tasks from a phone via Telegram while away
How to use it
- 1 Ensure Node 20 or later is installed
- 2 Run npm install -g agent-afk to add the afk command
- 3 Start tasks with prompts or skill commands such as a diagnose pipeline
- 4 Leave the desk; review tracking history and terminal states in the afk directory
Who it is for
Pros
- Free and open source, works with any model provider
- Verification workflow catches agents that claim false completion
- Reversible edits and isolated working trees limit blast radius
Limitations
- Command-line setup requires comfort with terminals and npm
- Remote supervision depends on Telegram, an external dependency
- Verification adds time to task completion
What agentafk is
Agent AFK is an open-source coding agent, licensed under Apache 2.0 and free to use, built around a specific insight: the bottleneck with coding agents is not speed, it is supervision. Developers start a task and then feel obligated to watch every step. Agent AFK is designed so you can genuinely leave, and it earns that trust architecturally rather than by promise. It is designed around six common failure modes of coding agents, baking orchestration, routing, tracking, verification, and supervision into the core. Sub-agent orchestration is on by default, with a single interrupt terminating the whole sub-agent tree. An adversarial verification workflow re-derives declared results before they enter difference comparison, so a claim of completion is actually checked. Tool calls appear inline with parameters, output, and timing, and every session is logged in a local afk directory. Remote supervision sends completion or stall notifications to your phone, where you can reply, redirect, or interrupt via Telegram.
What you can do with it
- Launch coding tasks and leave the desk, receiving verification receipts on completion
- Audit overnight or large refactoring runs through local tracking histories
- Reply to, redirect, or interrupt running tasks from your phone via Telegram
- Route prompts and skills like a diagnose pipeline running parallel hypotheses in isolated working trees
Who it is for
- Developers who want to focus on delivery rather than supervising code writing
- Developers handling large tasks where scrolling back through agent history is painful
- Developers using any model provider, including fully local models
- Teams that want auditable agent behavior with terminal states like completed, blocked, or asking
What to watch out for
Agent AFK assumes comfort with the command line: installation via npm, Node 20 or later, and reading local logs are all part of the workflow, which may deter developers who prefer GUI tools. Remote supervision depends on Telegram, an external service, so if that integration breaks or changes its API, your phone-based control loop suffers. Verification improves trust but adds time; fast iteration loops may feel slower than a plain agent. Even with adversarial checking, you remain responsible for reviewing changes before merging, since verification reduces but does not eliminate errors, especially on ambiguous requirements. Model costs still apply on whichever provider you connect, and running complex sub-agent trees can consume tokens quickly. Finally, as an open-source project it will evolve, so pin versions and read release notes before updating mid-project.
FAQ
Is Agent AFK free?
Yes. It is open source under the Apache 2.0 license and free to use, with any model provider you prefer, including local models.
How do I install it?
Install Node 20 or later, then run npm install -g agent-afk, which adds the afk command to your path.
What is the verification workflow?
An adversarial second agent re-derives the declared result before it enters difference comparison, ensuring a claimed completion is actually correct.
Why it is free
What the official page says： for free
Free limits, trial length and features are decided by the official site and can change at any time.
Product preview
Screenshot archived from the listing page, for preview only.
