Book a demo

Ticket in.
Merged PR out.Here's the receipt.

Four AI agents take a ticket to a merged pull request. You watch in plain English, answer when asked, and see what each ticket cost.

Get started Book a demo See how a run works

Runs on your serversWorks in your repoBills to your AI account

Troupe
RedPebble · Run receipt
RUN 041224 SEP 2026 14:02
TICKET #412
Add SSO
Time and model cost per agent for ticket 412
AGENTTIMECOST
Define4m$0.38
Developer22m$2.11
QA9m$0.71
Tech Lead3m$0.21
TOTAL38 min$3.41
QUESTIONS ASKED
1
APPROVALS
1
PULL REQUEST
#1187
0412 · 1187 · 3.41
Example runYour numbers will differ

How it works

Four agents.
One issue. One pull request.

Each agent does one job, then hands off. You step in only where you asked to.

  1. Define

    Reads the issue. Writes a short spec and the checks it must pass.

    Out: spec, 5 checks

    Gate: you approve

  2. Developer

    Writes the code on its own branch. Runs your tests.

    Out: commits, green tests

  3. QA

    Runs every check in a real browser. Keeps screenshots as evidence.

    Out: pass/fail, screenshots

  4. Tech Lead

    Reviews the whole change. Opens the pull request for you.

    Out: pull request

Configurable

Your team, your rules. Change who’s on it, teach them your conventions, and choose where they stop for your sign-off.

What you see

Follow every step
without reading code.

Each agent says what it did in one short sentence. When it needs you, it asks and waits.

Add SSO#412 · run 0412Example
38:00 elapsed$3.41
Define 4m
Developer 22m
QA 9m
Tech Lead 3m
  1. 14:02DefineReadthe issue and two linked threads.0:40
  2. 14:03DefineFoundSSO settings already exist for admins.1:10
  3. 14:05DefineWrotea spec with five checks.Approved1:50
  4. 14:06DefineHanded offto Developer.
  5. 14:07DeveloperReadthe login flow and session code.4:30
  6. 14:19DeveloperWroteOkta sign-in, behind a setting.11:40
  7. 14:26DeveloperRanthe test suite. 312 passed.3:20
  8. 14:28DeveloperHanded offto QA.
  9. 14:30QAAskedShould SSO users skip email verification?Waited 1m0:55
  10. 14:31YouYes. Skip it.
  11. 14:36QARanall five checks in a browser. All pass.4:40
  12. 14:37QAHanded offto Tech Lead.
  13. 14:40Tech LeadFoundnothing blocking. Opened PR #1187.2:50

The ledger

What a week of tickets costs.

12
tickets shipped
$41.08
total spend
1
question for you
Example week. What your AI provider charged, with no markup.
#TicketTimeCost
412Add SSOQ38m$3.41
409Fix CSV export timezone31m$2.87
415Rate-limit the public API54m$5.12
416Typo in invoice email22m$1.94
418Paginate the audit log47m$4.60
420Retry failed webhooks35m$3.08
+66 more tickets3h 41m$20.06
12 merged7h 28m$41.08

On-prem

Your servers.
Your repo.
Your AI account.

One deployment per client, inside your network. It can run with no internet at all. Nothing is mirrored to us.

YOUR NETWORK ACME CORP troupe DEFINE · DEV QA · TECH LEAD RECEIPTS · LOGS Your repo BRANCH PER RUN Model endpoint YOUR AI ACCOUNT You and your team APPROVE · ANSWER · MERGE BROWSER PUBLIC INTERNET OPTIONAL
  • Your networkTroupe runs on a server inside it: the four agents, their receipts and logs.
  • Your repoIt works in your repository, on a branch per run.
  • Model endpointIt calls the model with your credential.
  • You and your teamYou approve, answer and merge from a browser.
  • Public internetNot needed. Turn it off if you like.

Network

Air-gap ready.

Installs on your hardware. Runs with no outbound internet if you need it to.

Code

Stays in your repo.

Agents work on a branch in your repository. Nothing is copied off your servers.

Models

Your AI account.

Agents run on your own Anthropic or Kimi account. You pay the provider directly, and see every cent.

Before you ask

What people ask first.

Which code hosts and trackers does it work with?

Code on GitHub today. Tickets from GitHub Issues or Jira. Each one plugs in through the same connector, so adding another doesn’t change how Troupe works. Tell us what you use.

Which models does it use?

Claude through your own Anthropic account, or Kimi. You pick per team, and can choose a different model for each agent.

Who reviews the code?

You do. A Tech Lead agent reviews first, then opens the pull request. Nothing merges without your click.

What if an agent gets stuck?

It stops and asks you one question in plain English. The run waits until you answer.

What does it take to install?

One server you control. An install script brings it up with Docker Compose, then you connect your code host and your ticket tracker.

Watch it ship
one of your tickets.

See how a run works

Pick a ticket from your backlog. We'll run it with you, live, and you'll see what it cost.

Opens your email app with the request filled in.
READY TO SEND.

Your email app opened with the request filled in. Press send and we'll reply to pick a time.

Nothing opened? Write to troupe@redpebble.ai.

TROUPE · DEMO REQUEST