awayworkYou're away, your agents work

Run your coding agents on your real tasks, under rules you set. awaywork paces your quota and brings you only the decisions that matter.

Get started

How it works

GitHub

Your real tasks, picked by rules

Red CI, review threads, QA feedback, requested reviews and sprint tasks from your forge and task tracker, matched by an ordered queue of rules you control.

Skills, chains and plans

Skills are plain Markdown, built in or your own. Chain steps, loop review and fix, or split a feature into tasks that run as PRs, across repos too.

You decide what matters

Plans and posts wait for your reply on your own item, questions come to you, and a gate puts a decision card on a change that wants a human.

Vendor binaries, your login

Runs the stock agent CLIs (Claude Code, Codex, Cursor, more coming) headless with your own credentials. No API calls with a subscription token, no proxying.

Drafts, never merges

No forge token for workers. New branches and draft PRs by default, with opt-in for ready PRs or fast-forwarding your own PR's branch; awaywork continues its own draft only while it's where it left it, never force-pushes, merges or approves, and comments only on your items, marked as its own.

Paced quota

Stops each agent at a ceiling that spreads its quota over the window and keeps a reserve for you. Agents whose quota can't be read get fixed caps.

npm install -g awaywork
awaywork init             # opens Claude Code on /away, which sets it up with you
awaywork plan             # a dry look at the next run; only a tiny quota probe
awaywork schedule install # runs in the windows you set

Setting it up with a coding agent? Point it at llms.txt or With an AI agent.