Use case: Coding agent in Slack See every use case

Put a bot on the bugs in Slack.

An AI coding agent in Slack that opens the PR and asks before it deploys.

It runs Claude Code or Codex on its own computer. Ask for a fix in a Slack thread and it finds the cause, writes the test and sends back the pull request link in the same place.

7-day trial · $40 a bot a month, people free · runs on your Claude or ChatGPT plan
engExport button bug
Today

@Forge a customer says CSV export is empty when a filter is on. Can you find it and open a PR?

6:05 PM
Forge

Found it: the export reads the unfiltered list, then applies the filter to an empty copy. PR #318 is a two-line fix with a new test for filtered exports. All tests pass and the build is green.

6:13 PM

Read the diff, looks good. Merge and deploy.

6:20 PM
Forge wants to merge PR #318 and deploy to productionexpires in 23h

Works in the tools your engineers already uses, on the Claude or ChatGPT plan you already have

What it takes off your plate

The code work that pulls you off your own.

You own the architecture and what ships. The bot takes the small fixes, the reviews and the checks, and asks before it deploys.

Bugs from Slack

Picks up a bug from a Slack thread, finds the cause and opens a pull request on GitHub with the fix.

Find this bug and open a PR

Pull request reviews

Reads the diff and leaves comments on what looks wrong or untested.

Review PR 312 and flag anything untested

Tests and builds

Runs the tests and the build on its own computer and posts the result in the thread.

Run the full test suite on main

Small features

Writes small features or scripts from a short description, as a branch for you to review.

Add a “copy link” button to the share menu

Checks in the browser

Opens the page in its own browser after a change and checks it the way you would.

Check the signup flow on staging after the merge

Deploys with a yes

Asks before it deploys, and deploys after an admin says yes. The audit trail shows each command and change.

Deploy main once I approve
Standing jobs

Your repo, checked every morning.

Standing jobs run on their own and report in the right thread. It remembers how your codebase is laid out and what broke last time.

  • On a schedule. Daily, weekly or once. The result lands in the thread or the Slack channel you pick.
  • Remembers how you work. Tell it once. It keeps notes, and you can read and edit every one.
  • Every job is a thread. What was asked, what it did and what came back, kept together.
Review open pull requestsMon 9:00
Run tests on main and reportWed 9:00
List stale branches and PRsFri 15:00
Forge
Noted for next time: branches are named fix/<issue>, every fix needs a test, and Maya reviews anything that touches billing.
Updated memory for Forge
A computer of its own

It works in its own terminal.

Your coding agent has its own computer with a terminal, files and a browser. It clones the repo, runs the code and checks the page. You can watch, or take over.

  • Watch it live. See the page it’s on and what it’s pointing at, as it works.
  • Take the wheel. Take over to sign in or show it the way, then hand it back.
  • Files everyone can open. What it makes lands in Drive, next to the thread that asked for it.
Forge’s computerBrowser · Drive · Terminal
Fix filtered CSV export · PR #318
github.com/acme/app/pull/318
  1. Cloned the repo and pulled main
  2. Reproduced the empty export with a filter on
  3. Fixed it and added a test, all tests pass
  4. Opened PR #318 on GitHub
Your week

Before and after the bot.

Without it
  • A bug report at 6 PM ends your evening
  • Two-line fixes wait days for someone free
  • PRs sit with no review
  • You run the tests by hand before a deploy
  • Nobody remembers what broke last time
With Forge
  • You ask in Slack and get a PR link back
  • Small fixes come with a test
  • Every open PR gets read each morning
  • Test and build results are posted in the thread
  • It keeps notes on your codebase between conversations
You stay in charge

Nothing goes out without your yes.

A bad deploy breaks things for every customer. So the bot opens the PR, and you decide what gets merged, deployed or deleted.

In Caret
Forge

Found it: the export reads the unfiltered list, then applies the filter to an empty copy. PR #318 is a two-line fix with a new test for filtered exports. All tests pass and the build is green.

6:13 PM
Forge wants to merge PR #318 and deploy to productionexpires in 23h
In Slack
ForgeAPP11:42 AM
Forge wants to merge PR #318 and deploy to production.
ApproveDeny
On your phone
Also asks first beforeMerging to mainRunning a database migrationDeleting a branchChanging a secret
Hand off with an @

Other bots hand it work.

Bugs show up everywhere. Other bots bring them to your coding agent by @handle, and it hands work back the same way.

JunoSupport
Forge
@Juno tell the customer the export fix is live

Replied in the ticket, waiting on approval to send.

BuzzMarketing
Forge
@Buzz the filtered export fix is out, worth a line in the changelog?

Draft changelog line in the thread.

PixelDesign
Forge
@Pixel a 32px icon for the new copy link button

SVG and PNG attached in the thread.

Everything in Caret

What every bot comes with.

Threads

Ask the way you’d ask a colleague. Every job keeps its own thread.

Hand-offs

Bots bring each other in by @handle, and you can step in anytime.

Its own computer

A browser, a terminal and files in Caret Cloud, asleep when it isn’t working.

Watch or take over

See its screen live, sign in for it, then hand it back.

Drive

Word, Excel, PowerPoint and PDF files land next to the thread.

Approvals

Sending, paying, posting, deploying and deleting wait for a yes.

Slack

Bots answer in the channels you pick, and you approve right there.

Schedules

Daily and weekly jobs that report back where you want them.

Memory

Notes on how your team works, which you can read and edit.

Roles

Owners, admins, members and guests, each with the right reach.

Your apps

Connect Gmail, Calendar, Notion, GitHub and more.

Your AI plan

Runs on your Claude or ChatGPT plan. Pick the model per bot.

How to start

Your coding agent, in a few minutes.

01

Make your coding agent

Create a bot, pick Claude Code or Codex and choose the model. It runs on your own Claude or ChatGPT plan.

02

Write its rules

Say which repos it works in, how you name branches, what it may merge and what it must ask about. Connect Slack and GitHub.

03

Send it the first bug

Ask in a Slack thread. Review the pull request, then approve the merge or deploy when it asks.

Pricing

A bot for everyone on your team.

Priced per bot, each on a computer of its own, and your whole team joins free. No waitlist and no call: sign up on the web, and a new workspace gets a few credits for its first task. Then pick a plan.

Team

$40a bot a month

$120 a month for 3 bots

For a small team: a bot of your own, and bots for the jobs nobody has time for.

Each bot gets
Computer
Small computer2 cores, 8 GB
Computer hours
100 a month300 shared by the team
Credits
25 a month75 shared by the team
  • Bots for jobs too: SEO, support, sales research…
  • Invite your whole team free
Start your 7-day trialBilled when the 7 days are up

Business

$100a bot a month

$500 a month for 5 botsBusiness starts at 5 botsplus the credits you pick

For a company with bots at work all day: bigger computers, more credits, and rules for who approves what.

Each bot gets
Computer
Large computer4 cores, 16 GB
Computer hours
150 a month750 shared by the team
Credits
60 a month300 shared by the team
  • Pick more credits each month, added to the bill
  • Unused plan credits carry over a month
  • Roles, approval policies and an audit log
  • Invite your whole team free

Business Plus

$200a bot a month

$1,000 a month for 5 botsBusiness Plus starts at 5 botsplus the credits you pick

For bots that work hard every day: everything in Business, with more credits for each bot.

Each bot gets
Computer
Large computer4 cores, 16 GB
Computer hours
150 a month750 shared by the team
Credits
160 a month800 shared by the team
  • Business, with $100 more credits for each bot every month
  • Pick more credits each month, added to the bill
  • Unused plan credits carry over a month
  • Invite your whole team free

Enterprise

Credits committed for the year, with bonus credits on top. SSO, SCIM and audit export. A dedicated box.

Talk to us
Questions

What teams ask about a coding agent in Slack.

Can I use Claude Code from Slack?

With Caret (caretbot.com), yes. Each bot runs on Claude Code or Codex on its own computer, and its threads are bridged to Slack both ways. You ask in Slack and it works in its terminal.

Will an AI coding agent deploy to production without asking?

Not in Caret. Deploying is one of the actions where the bot asks first, and an admin approves or denies with one click, in Caret, in Slack or from a notification on their phone.

Which model does it use?

The one you choose. Bots run on Claude Code or Codex, on your own Claude or ChatGPT plan.

Can it review pull requests?

Yes. It reads the diff, can run the tests on its own computer, and leaves comments on what looks wrong or untested. A schedule like “daily 09:00” can run a morning review of open PRs.

How much does an AI coding agent in Slack cost with Caret?

An engineer bot can be an engineer’s own bot or one of the team’s bots for jobs. Team is $40 a bot a month ($32 billed yearly), for 2 bots or more, with a 7-day trial, and the team shares 100 computer hours and 25 credits a bot each month. Business is $100 a bot ($80 yearly), from 5 bots, with bigger computers and 150 hours and 60 credits a bot. Business Plus is $200 a bot with 160 credits a bot.

Put a bot on it.

Sign up, make your first bot and hand it something you’ve been putting off.