agentafk

A practical workflow

1. Install Agent AFK via npm, set your API key, and run your first command in under two minutes.
2. Start a task in the terminal and walk away; the agent verifies its own work and notifies you when done, providing a receipt of every step.
3. Use slash commands like /review, /mint, /diagnose, and /ship to orchestrate complex workflows with parallel subagents and verification steps.

Based on the sources below. Editorial review does not imply hands-on product testing.

Candidates based on category and primary feature. Check feature and pricing differences before switching.

What changed

Changes to the facts recorded here, not a live scan of every vendor update. Save this tool to follow updates in your account.

1. 9/13/2026

   Updated: best for, export formats, features, limitations, platforms, summary, use cases, walkthrough

   See recorded changes

   bestFor

   Before: ["Developers who want to run AI coding agents autonomously without constant supervision and receive notifications when tasks complete.","Engineers running headless or CI-adjacent workflows needing model reasoning, bash access, and subagent orchestration without human intervention."]

   After: ["Developers who want to hand long or multi-step AI tasks to an agent and check in asynchronously without babysitting. Engineers running headless or CI-adjacent workflows needing model reasoning, bash access, and subagent orchestration without human intervention.","Users who want Telegram supervision to receive notifications on their phone when tasks finish, without needing to set up a server."]

   exportFormats

   Before: ["Local traces are stored as JSONL files under ~/.afk/ for inspection and deletion."]

   After: []

   features

   Before: ["Subagent orchestration allowing multiple parallel or dependent tasks with one interrupt killing the entire subagent tree.","Verification workflows with adversarial verification to ensure task completion correctness before finalizing changes.","Remote supervision via Telegram bot that notifies users when tasks finish or stall and allows remote interaction.","Local trace history stored under ~/.afk/ providing detailed logs of every session, subagent, and tool call for inspection.","Safety and verification controls including cost ceilings, turn limits, path-access approval, and worktree isolation to keep unattended runs safe."]

   After: ["Subagent orchestration allowing many subagents to run in parallel or sequentially as needed.","Verification workflows where a second adversarial agent re-derives claims before results reach the user, ensuring correctness.","Telegram supervision that sends notifications to the user's phone when a run finishes or stalls, enabling remote monitoring and control.","Local traces that append inspectable logs of tool calls, decisions, and costs for every run, viewable with a trace command.","Safety and verification controls including cost ceilings, turn limits, path-access approval, and worktree isolation to keep unattended runs safe and bounded."]

   limitations

   Before: ["Requires Node.js version 20 or higher to install and run the CLI."]

   After: ["Requires Node.js version 20 or higher to run.","Not fully autonomous; it stops and asks the user if unsure, requiring some supervision in ambiguous cases."]

   platforms

   Before: ["Runs locally on user machines with Node.js environment; supports CLI, daemon, and Telegram bot interfaces.","Four surfaces sharing one session manager: One-Shot Chat, Interactive REPL, Daemon, and Telegram Bot."]

   After: ["Runs locally on Node.js environment (Node ≥20) with CLI, daemon, and Telegram bot interfaces sharing one session.","Four surfaces: CLI one-shot chat, interactive REPL, daemon for unattended runs, and Telegram bot for mobile supervision."]

   summary

   Before: "Agent AFK is an open-source local AI agent runner designed to autonomously run coding tasks without supervision, providing verified receipts and remote notifications. It supports multiple interaction surfaces including terminal, daemon, and Telegram bot, sharing one session and context."

   After: "Agent AFK is an open-source local AI agent runner that executes coding tasks autonomously while the user is away, verifying its own work and notifying the user upon completion. It operates through multiple interfaces including a terminal CLI, daemon, and Telegram bot, sharing a single session and local state without sending data externally."

   useCases

   Before: ["Running long or multi-step AI coding tasks autonomously without needing to watch the process continuously.","Supervising AI agent runs remotely via Telegram notifications and interacting with the agent from a phone."]

   After: []

   walkthrough

   Before: ["Install Agent AFK via npm, set your API key, and run your first command in under two minutes.","Use slash commands and skills like /diagnose and /review to run parallel checks and verification workflows.","Receive Telegram notifications when tasks finish or stall and respond remotely to control the session."]

   After: ["Install Agent AFK via npm, set your API key, and run your first command in under two minutes.","Start a task in the terminal and walk away; the agent verifies its own work and notifies you when done, providing a receipt of every step.","Use slash commands like /review, /mint, /diagnose, and /ship to orchestrate complex workflows with parallel subagents and verification steps."]