shipgremlins. Star on GitHub

OPEN SOURCE. SLIGHTLY FERAL.

AI product managers
that actually
use your app.

Your app has a backlog. And blind spots. Give PM Gremlins a goal worth pursuing; they explore, investigate, and propose improvements. Approve the direction. Your crew codes, tests, and brings you a batch of improvements to review.

Self-hosted · Real browsers · Your final say

MEET THE NIGHT SHIFTGREMLIN / 001
A mischievous lime green gremlin holding a browser bug report
“Found something.”Of course you did, little buddy.
001
Will absolutely click that button twice.

WHEN TO CALL IN THE GREMLINS

You built the app.
Give it a crew.

The onboarding you keep putting off. The feature nobody has time to explore. The permissions that need a closer look. Start with one part of your existing product that deserves more attention.

  1. 01 / CONNECT YOUR APP

    A home. A job. A gremlin.

    Connect your repository. A Setup Gremlin reads the code and suggests a crew with distinct responsibilities. Review their briefs, connect Linear, and test access to the app before starting patrols.

    Start with one neglected area.
  2. 02 / APPROVE THE DIRECTION

    “Make the first import easier.”

    Your PM investigates and proposes an epic: an outcome, boundaries, and a definition of success. Approve that scope once. It breaks the work into testable tickets for Coding Gremlins.

    Product ideas with something behind them.
  3. 03 / REVIEW A TESTED BATCH

    A batch worth your attention.

    Coders implement; PMs test each change. Failed checks return with feedback. Each PM collects its passing work into a promotion PR, at 10 tickets by default. You review the batch into staging.

    Your runners. Your release decision.

Useful before launch. More useful as the app grows. Bring a working prototype or an established product. You don’t need a big team; you need a real codebase, a direction, and time to review the results.

Try one improvement
EARLY ALPHA / BRING YOUR APP

Adopt your first gremlin.

Node 22.12+ · Git · Docker · Your first adoption
ONE INSTALL. A WHOLE CREW.
$ npm install -g git+https://github.com/AgentBurgundy/shipgremlins.git
$ gremlins setup

Then connect your app in the private dashboard and give your first gremlin a useful job. See the next steps →

BUILT AROUND YOUR WORKFLOW
GitHub
GitLab
Vercel
Linear
Slack
Your tools. Local workers.

A LITTLE CHAOS.
A LOT OF SCREENSHOTS.

Your app's
new night shift.

Give them a mandate.
Let them poke around.
Review what they find.

MEET YOUR TWO CREWS

One spots the gap.
One closes it.

Different jobs. Shared evidence. PM Gremlins investigate and QA. Coding Gremlins implement the milestones inside your approved epic.

THE INVESTIGATORS

PM Gremlins

Give them a mandate. They explore the app, propose bounded epics, and test each implementation against its acceptance criteria.

  • Real browser exploration
  • Product ideas grounded in your app
  • QA evidence and a promotion batch
Their handoff: improvements worth shipping.
PM gremlin investigating a shield with a magnifying glass
THE BUILDERS

Coding Gremlins

They pick up eligible tickets from approved epics on your runners. The controller checks and merges internal PRs into the test branch.

  • Scoped implementation
  • Configured checks before integration
  • Bounded fixes from PM feedback
Their handoff: code ready for PM testing.
Coding gremlin showing a working app on a phone

Launch either crew from your dashboard. Follow visible tool calls, progress, checks, and screenshots in Activity.

01 / THE PM MANDATES

A particular set
of nitpicks.

One PM owns one area. Give each a mandate, a schedule, and boundaries. Let curiosity do the rest.

THE SUSPICIOUS ONE
Security gremlin inspecting a shield and lock with a magnifying glass

The Security PM

“Should that user really be able to do that?”

  • Roles and permissions
  • Cross-tenant boundaries
  • Auth and session edge cases
THE NOSY ONE
Feature gremlin examining an unusually long CSV spreadsheet

The Feature PM

“What are they trying to do next?”

  • Unmet goals and repeated work
  • Alternative ways to solve a problem
  • Small, testable feature ideas
THE PICKY ONE
Experience gremlin inspecting a button on a mobile phone

The Experience PM

“It works. But does it make sense?”

  • Signup and first-run friction
  • Mobile layouts and accessibility
  • Confusing flows and dead ends

Your mandate. Your codebase. Give each PM an outcome, run discovery, and let it build a feature map and ranked ideas for what comes next. Shared project decisions keep the crew working toward the same goal. Meet the PM workspace →

02 / SHOW YOUR WORK

“Trust me”
isn’t a test.

Every useful finding needs a reproduction. Every verified fix needs evidence. Follow the trail from a browser observation to a production merge.

Illustrative workflow. No live agents are running on this page.

MANDATEKeep every tenant in its lane.EXAMPLE RUN
01
OBSERVE

A member opens another team's export.

Reproduce with two isolated test accounts. Record the request and the actual browser state.

02
APPROVE THE EPIC

A bounded outcome. A plan to test it.

Review the epic’s scope and evidence in your project inbox. Your PM turns the approved direction into testable tickets.

03
IMPLEMENT & VERIFY

The owner can export. The other team can't.

The PM checks the deployed revision and acceptance criteria. Passing work joins its batch; a failure returns to coding with evidence.

Verified is a milestone. Production is Done.You review the promotion batch into staging.

03 / GUARDRAILS INCLUDED

Ambitious agents.
Sensible boundaries.

Your crew can be relentless without having the keys to everything.

01

A mandate, not a blank check.

Control the epic’s scope, test environment, and amount of work in flight.

02

Break. Diagnose. Repair.

Bounded retries and a dedicated recovery lane keep ordinary work paused while a broken integration is repaired.

03

No evidence, no promotion.

Require current deployment evidence before a promotion opens. Stale checks and a confident comment don't count.

A CHANNEL FOR THE CHAOS

Your gremlins
have entered
the chat.

Bring PM findings and Coding Gremlin updates into Slack. Pick a workspace and channel from your dashboard, then keep up with the work where your team already talks.

Set up Slack

One channel for both crews. Human approval and production decisions stay with you.

# gremlin-updatesILLUSTRATIVE EXAMPLE
PM GremlinAPP

Found something worth a look.
A member can open another team’s export. Reproduced with two test accounts; screenshots are attached to the finding.

PM FINDING

Security mandate · example-app
Epic ready for owner review

Coding GremlinAPP

Implementation is ready for PM QA.
The export fix passed its code checks. The controller will integrate it before the PM tests the deployed acceptance criteria.

CODING UPDATE

Epic milestone · example-app
Internal PR prepared. Production unchanged.

Sample messages show the intended handoff. This page is not connected to a live Slack workspace.

04 / BRING YOUR STACK

Your tools.
A few new teammates.

Keep your code, tickets, and infrastructure.
Give your PMs the evidence behind the screen, too.

AVAILABLE IN THE ALPHA

Your machine. Your crew.

Docker runs the agents on your computer or homelab. Connect GitHub or GitLab for source and PRs or MRs. Linear tracks the work.

Local Docker workers · Real browsers · Claude Code
OPTIONAL PROJECT CONNECTIONS

Your PMs can read the logs.

Sentry logs and errors. Datadog logs. Mixpanel analytics. Each PM can inspect its project's failures and usage, then connect those signals to what it sees in the browser.

Project scopes · Read access · Saved Insights reports

Connect your project's signals

FROM YOUR DASHBOARD

Set up once. Keep improving.

Follow the next setup step. PMs prepare missing Linear projects and labels. Connect the test preview and a dedicated app account, verify your worker, then activate the ready crew.

No fork · No automation repo · No provider CI required
A place for every mandate.Connect Linear. Your PM prepares missing mappings and labels. Check the actual preview.Connect Vercel, select the preview, then test app sign-in. Your Railway environment.Connect a token and discover the existing service you want tested. A test home for every crew.Let a Setup Gremlin recommend hosted staging or a disposable Docker app.

THEY'RE SMALL. THEIR STANDARDS AREN'T.

Start with one area.
Meet your first gremlin.

Bring the app you care about. Pick the flow that needs attention. Approve a useful direction. Review the batch when it is ready.

Adopt your first gremlin
YOUR TERMINALNode 22.12+ · Docker

# Bring the crew home

npm install -g git+https://github.com/AgentBurgundy/shipgremlins.git

# Open your private setup dashboard

gremlins setup

# On your server or homelab · port 4311

gremlins setup --lan --port 4311
Read the setup guide

Early alpha, built in the open. Bring Node.js, Git, Docker, and your provider credentials. Workers run on your machine; you approve the work and control production merges.

A FEW FAIR QUESTIONS

Before you let
them loose.

When does it make sense to use ShipGremlins?

When you already have an app and more improvements than you have time to investigate and implement. Start with one outcome: a clearer first-run experience, a missing workflow, or better account isolation. The crew investigates and implements approved work; you set direction and review the result. Choose your starting point →

Will it build my app from an idea?

There is an experimental idea-to-app path. It creates a private-by-default repository, proposes a small crew, and lets you review a foundation coding assignment. Expect to iterate on the result: a scaffold is not proof of a polished, integrated product. Improving an existing app is the main workflow. Read the experimental guide →

Does my app need to be launched already?

No. A real codebase and a clear goal are enough to start repository discovery. Browser investigations need a runnable app in an isolated test environment; a local prototype can use supported Docker setup. Add Linear when you’re ready for proposals and approved coding work. Start with one assignment →

Can I retire a project or a gremlin?

Yes. Delete projects and PMs from the dashboard, with a clear preview and a local recovery copy. Restore them from Settings with automation paused. Your repositories, Linear tickets, and past runs stay intact. Removal and recovery →

What is a PM Gremlin versus a Coding Gremlin?

PM Gremlins investigate a mandate, propose epics, and QA implementations. Coding Gremlins implement eligible tickets within an approved epic. Internal PRs go through automatic checks and integration; each PM batches its verified tickets into a promotion PR. You approve epics and review batches into staging. Existing projects keep their configured delivery policy.

Does every project need hosting?

No. Add your GitHub or GitLab repository and let the Setup Gremlin recommend a test home. Use an existing staging environment or a disposable Docker app with optional PostgreSQL and Redis. Review any proposed setup files as a draft PR or MR, then test access in a real browser. Libraries and CLIs can stay in repository-only mode. See project onboarding →

Can my crew report to Slack?

Connect a workspace and channel in your self-hosted dashboard. When one-click installation is available, Add to Slack guides you through it. An advanced incoming-webhook option is also supported. Slack carries updates; it does not approve epics or merge releases.

Do I have to wire up every Linear project by hand?

No. Save a Linear connection and select it for your app; the PM prepares missing team, project, and label mappings when it needs them. Existing mappings stay in place. Inspection, adoption, and the first read-only Discovery need no Linear mapping. New PMs start paused. If workspace permissions or limits block setup, your PM stays saved for retry.

Do I have to review every ticket and internal PR?

New managed projects ask you to approve a bounded epic and review each PM’s promotion batch, at 10 verified tickets by default. The crew implements and tests the child tickets; the controller checks and merges internal PRs into pm-staging. Failed QA returns with feedback and bounded retries. Existing projects keep their selected approval policy. Follow the delivery loop →

Does connecting Vercel let gremlins sign into my app?

It opens the protected preview. Private app features still need a dedicated test account. Setup can suggest the password login controls from source, then checks the actual login with your saved credentials. You can choose public pages only, with private journeys clearly left untested. Set up app access →

Does it ship to production by itself?

No. Agents work on approved scope in a test environment. Promotion has verification gates, and the owner retains staging and production merge decisions. Reconciliation marks work Done only after production inclusion is proven.

Can I run it on my own server?

Yes. Run gremlins setup --lan --port 4311 on your server, then open the private session link from a browser on your trusted network. Create local Docker workers in the dashboard. Connections and configuration live on that server; jobs receive the credentials they need. No fork, automation repository, or GitHub Actions/GitLab CI runner registration is needed.

Can workers run on other machines?

Yes. Enroll a remote Docker worker from the dashboard, choose its projects, and run the one-time connection command on that machine. Workers connect outbound and use expiring job leases. A cloud VM, Linux server, or Mac/Windows Docker Desktop host can provide capacity. Native iOS jobs and automatic cloud provisioning are still planned.

Which AI models can I use today?

The current workers use Claude Code. Additional runtimes are planned. You control the credentials and model configuration in your own environment.

Does it work with any web app?

The browser can explore many kinds of apps, but reliable testing needs configured accounts, isolated data, and acceptance criteria. GitHub and GitLab source repositories are supported. Use Vercel, Railway, Cloud Run, or a reachable preview URL for browser work. Provider connectors discover existing deployments; they do not deploy or provision your app.

Can it create test images and CSVs?

Yes. The fixture CLI generates CSVs from JSON and reproducible PNG test images, ready to upload through Playwright MCP. Semantic AI image generation and automatic test-data cleanup are on the roadmap.