AgentCode
Download

Stop babysitting your agents.

AgentCode tells you which one needs you, keeps the rest working, and switches accounts at the limit. From devs, for devs.

Download for MacIntel Mac

Free. No AgentCode account. No servers of ours.

Needs you 5: a Write, two Bash commands and an Edit to allow, and a question, across api-gateway, storefront and docs, each with minutes left
Usage per account: Work at 100% of its 5-hour session, Personal at 27%, ChatGPT at 8%

Ten tabs. One is stuck on Allow?

WhenIn terminal tabsIn AgentCode
An agent waits for approvalIt sits there until you find its tabIt tops Home under Needs you
An account hits its limitWork stops until the resetIt moves to your other account
You log off for the nightThe rest of your plan goes unusedSmart agents run on what is left
You step away from the deskYou find out when you are backYou answer from your phone

Come back to what finished

Nine agents finished while you were in another tab. Home keeps them listed until you look.

Finished since you looked 9: Gateway docs, Checkout refactor, Load test and Release prep lead, across api-gateway and storefront

Read it, then answer in one click

An Edit diff turning > into >= in shipping.js and the Allow Edit card, waiting for you
Read the change, then Allow or Deny. Nothing is written first.
The agent has a question: which format for the 0.8 release notes, with three options and a free answer
Pick an option or type your own. The agent carries on.
AgentCode on a phone: the Allow Edit card under its diff
The same card on your phone, over your own Tailscale network.
Three messages queued above the box while the agent works, sent in order after this response
Type while it works. Messages wait their turn, in order.

Your agents don’t stop when you do.

Close the window and AgentCode keeps running in the menu bar.

  1. At the limit, it switches

    Work hits its 5-hour cap. The agent moves to your next account and keeps going.

    When an account hits its limit: switch across providers, within this provider, or wait for the reset

  2. Spare usage gets spent

    Smart agents wait until you are away and your plan has room, then start.

    Smart agents: Dependency audit, waiting until you are away

  3. Mornings start done

    A routine runs your prompt on weekdays at 9:00, before you sit down.

    The routine’s schedule: weekdays at 9:00, next on 26 Oct, a fresh agent each run, skipped if the last one is still running

One agent hands work to others

The agents list: Release prep lead with two workers nested under it, one done and one needing approval
Each worker gets its own branch. The lead reviews, then merges.

See every feature

What it asks of you

Installs
Nothing first. The Claude runtime ships inside the app; Codex needs one codex login.
Cost
Free. No account with us, no email.
Your logins
Your own Claude plan or API key, and your own ChatGPT for Codex.
Your data
Projects and transcripts stay in ~/.agentcode. Keys go in the Keychain.
First launch
Signed and notarized by Apple. It opens with a double-click.
Leaving
Drag it to the Trash. Your repos are untouched.

Stop hunting for the tab that needs you.

Version 0.7.2, released October 26, 2026, for Apple silicon and Intel. Updates arrive inside the app.

Download for MacIntel Mac
A worker asks to run npm test; Always for Release prep lead keeps the rule npm test * for the lead and its workers