Anthill

Design the system behind the prompt.

You already write the whole loop by hand — who does what, what gets reviewed, what happens when tests fail. Anthill lets you draw it instead, hands you the prompt, and then shows you what your agent is actually doing.

WorkflowCheckout flowFrom CodexReady for agent
Waiting for a session to hand over a workflow
CodexTest Projects
Build the checkout flow. Have one agent design it and a second one attack the design before anything is written. Then implement it, run the suite, and loop back on failures until the tests pass. Ask me before changing any public contract.
▍
GPT-6 Sol High
Works with
Claude CodeCodexVS CodePi
Watch it happen

Stop re-reading the transcript to find out where you are.

Whichever way it started, Anthill recognises the session and lights up your workflow as the work moves through it. Which step is running, which are finished, how long each one took, which one is waiting on you — at a glance, not by scrolling back.

StartAGENT STEPdoneImplementDeveloper · 1m 29sRUN TESTSRun testsWorking · 42s so farAGENT STEPqueuedFix failuresDeveloperrunningdoneneeds youfailedqueued
Activitynewest first
T
Testercalled inRun tests14:26:19
BashRun the project’s test suite◌Running
T
Testerstarted inRun tests14:26:02

Run the suite and report the real result, including the names of any failures.

Running…
Claude Codecalled inOrchestration layer14:25:58
WebFetchFetch the changelog page?No result
Claude Codesaid inOrchestration layer14:25:51

Starting on the limiter. I will put the counter behind a small interface first.

Anthill reads the records the CLI already writes on your machine. No transcript, no file contents, and none of the model's private reasoning.

See the workflow while your agent works.

To follow a run step by step, end your prompt with the watch command: /anthill:workflow watch in Claude Code or VS Code, or $anthill watch in Codex. The agent starts right away, and Anthill opens the workflow and lights up each step as it runs.

In the Claude Code app, end the prompt with /anthill:workflow watch.
Search
New
Artifacts
Routines
Anthill
Test Projects
Ungrouped
Checkout flow
Anthill workflow setup
Parity test failures
ANT-86 desktop QA workflow
NNNick· Max
My AppMy App

The cart already stores line items per customer. I’ll keep the checkout logic in its own module.

Handing the workflow to Anthill and starting on Implement.
Anthill is following this sessionCheckout flow · 4 steps · live
My Appmain+214 −18
Build a checkout flow, test it, and fix failures. /anthill:workflow watch▍
AutoOpus 5High
The plugin opens Anthill for you.The workflow appears on its own, and each step lights up as your agent reaches it. Nothing to copy, nothing to start.
Live sessionCheckout flow1 of 4 steps done · 18m 54sWatching · Claude Code
tests failedStartAGENT STEPdoneImplementDeveloper · 18m 12sRUN TESTSRun testsWorking · 42s so farAGENT STEPqueuedReportDeveloperAGENT STEPqueuedFix failuresDeveloperDone
runningdoneneeds youfailedqueued
Activitynewest first
T
Testercalled inRun tests14:26:19
BashRun the project’s test suite◌Running
T
Testerstarted inRun tests14:26:02

Run the suite and report the real result, including the names of any failures.

Running…
D
Developersaid inImplement14:25:51

Done with the checkout. Tests are next.

Claude Codesaid inOrchestration layer14:07:39

Starting on the checkout. I’ll keep the cart logic in one module.

Install

Get Anthill, then add it to your coding tool.

Anthill is free and open source. Install the app, add the plugin to Claude Code, Codex or VS Code, and tag Anthill in a prompt. It displays and follows the work. It never starts, stops or steers your agent.

Get the app
macOS desktop appSupported

The desktop app, with the canvas and the Live session. On first launch it finds Claude Code and Codex and offers to add its plugin to each, and shows the one setting VS Code needs.

Add the plugin

The macOS app adds it to Claude Code and Codex for you on first launch. On Linux and Windows, or if you skipped that step, add it with your tool's own commands below. VS Code has no command for plugins: two links it opens itself add Anthill's marketplace and install the plugin. It needs Node.js. Start a new session afterwards so it loads.

Claude Code

Works in the Claude Code app and in the terminal. Call it with /anthill:workflow design or watch.

$ claude plugin marketplace add anthillapp/anthill
$ claude plugin install anthill@anthill
Claude Code plugin guide
Codex

Works in the Codex app and in the terminal. Call it with $anthill design or watch.

$ codex plugin marketplace add anthillapp/anthill
$ codex plugin add anthill@anthill-local
Codex plugin guide
VS CodeBeta

Works with VS Code's agent, in the Agents window and in Agent mode. Call it with /anthill:workflow design or watch.

To install it, run these two commands and confirm each in VS Code: the first adds Anthill's marketplace, the second installs the plugin from it.

$ open "vscode://chat-plugin/add-marketplace?ref=anthillapp/anthill"
$ open "vscode://chat-plugin/install?source=anthillapp/anthill&plugin=anthill"

On Windows, use Start-Process instead of open in PowerShell; on Linux, xdg-open. If VS Code does not open, the plugin guide has the same steps by hand.

VS Code plugin guide
CursorComing soonA plugin for Cursor’s agent is in progress.
Source
github.com/anthillapp/anthill · Free and open source. Workflows are ordinary JSON files you can read, diff and commit.Open the repository

Not sure where to start? Read the README.

Where it pays off

Built for the long jobs — many steps, many agents, hours of work.

A one-line fix needs none of this. A migration, a refactor across packages, a feature that has to be designed, reviewed, built and tested does — and those are exactly the jobs where a paragraph of prose stops being enough and the agent disappears for an hour.

PlanBuildTestsFixShiptests failedone long paragrapha map you can see

Many steps

Ten or twenty stages with reviews and loops between them stay readable as a diagram, and stay editable next week.

ADRT?ADRTplancodepassedfixArchitectDeveloperReviewerTesterwho is doing what?clear hand-offs

Many agents

Specialists that hand work to each other, argue with each other, and come back to their own work later.

1h 40mPlan4mBuild52mTests9mReviewnowDeployReview needs your answerscroll and guessyou are here

Long runs

When a job takes an hour, the question is not what the agent said — it is which step it is on, and whether anything is waiting on you.

Questions

Does Anthill run my agents?

No. It plans the work and then watches. The prompt it produces is an instruction for Claude Code or Codex, and the ordering, conditions and limits in it are for that tool to honour — Anthill does not start, stop, pause or answer a session, and turning off observation leaves a running session untouched.

How does it see what is happening?

It reads the records the CLI already writes on your machine, and it says how confident it is about each one. An event it cannot tie to a step is shown as unmapped rather than guessed at. Your workflows and sessions never leave your computer.

Do I need an API key?

No. Anthill borrows a coding CLI you have already installed and signed into. There is nothing to connect and no key to paste.

How do I hand a task to Anthill?

Install the Anthill plugin for Claude Code or Codex, then end your prompt with /anthill:workflow design in Claude Code, or $anthill design in Codex. Use watch instead of design to skip the review and follow the work live. Prefer to draw it first? Design it in Anthill and copy the prompt.

Is it really open source?

Yes, and free. Workflows are ordinary JSON files on disk, so one you build is yours to read, diff and commit next to the code it is about.

What do I need to run it?

A Mac, and Claude Code or Codex installed and signed in. Anthill finds them on first launch and installs its plugin for you. On Linux it runs from source as a local web app; Windows works from source as an experiment.