Customer feedback in. Merged pull requests out.
Starko is a product factory. It collects what customers say in support conversations and session replays, turns it into sized, costed recommendations, lets a coding agent build the ones you accept, and verifies the result before you merge. You pay per merged pull request. Click around the dashboard below.
Improvements
What Starko AI found in your visitors’ sessions, what you decided, and what it cost.
Accepting a row starts a run with the credit cap shown. Declining costs nothing.
Proposed · 30 days
23
Accepted · 30 days
17
Shipped · 30 days
14
Declined · 30 days
6
Credits spent
1,840
Visitors spared
412
Inputs to the factory: every support channel you already use, session replays from your product, and the Starko widget on your site.
- Slack
- Viber
- Telegram
- Messenger
- Intercom
- Discord
- Linear
- GitHub issues
Three strokes.
Three stages.
Most of what customers tell you never becomes a change. Starko runs every message, email, Slack thread and rage-click through the same loop: collect it, decide whether it is worth building, ship it as a merged pull request. The mark is that loop drawn once.
- 01
Collect
184 conversations · 41 session replays this week
WhatsApp, email, Slack, Instagram and the website widget land in one inbox. Starko AI answers what the knowledge base covers and files a ticket for the rest, linked to the customer and company.
- 02
Decide
23 recommendations · 17 accepted · 6 declined
Repeated friction becomes a recommendation: bug or improvement, Small, Medium or Large, with the visitors behind it and a credit cap. You accept or decline each one. Nothing is built until you do.
- 03
Ship
14 merged pull requests · 412 visitors spared
An accepted recommendation becomes a run: a coding agent works in a sandbox, a reviewer agent checks the diff, and a pull request with evidence waits for your merge. Merge closes the ticket.
How the factory works Four stages, one loop, a human at the merge button.
Starko does not replace your coding agent; it operates it. Claude Code, Codex or Cursor writes the code. Starko decides what to build, writes the brief, runs the sandbox, reviews the output and keeps the ledger.
Collect
01Every channel and session replay
Recommend
02Sized, costed, tied to customers
Build
03A coding agent in a sandbox
Verify
04Tests, evidence, reviewer agent
- Test
- Evidence
- Review
Intake
and channels
Conversations from WhatsApp, email, Slack, Instagram and your website widget become tickets with a customer, a priority and a type. Starko answers what it can from the knowledge base and routes the rest to the right project.
Learn more about intake and channels- lelena 10:52 AM
Anyone else seeing the WhatsApp template picker crash when a template has no body? Two customers hit it today.
- dididier 10:53 AM
Yes, Acme reported it too. Picker also lists 41 stale templates; there is no way to delete them.
- enena 10:55 AM
That is three customers this week. Can we get this into the factory before Friday?
Recommendations
and decisions
Repeated friction becomes a recommendation: bug or improvement, sized Small, Medium or Large, with the visitors behind it and a credit cap. You accept or decline from Home or the Improvements page, and nothing is built until you do.
Learn more about recommendations and decisionsRecommendation: make the search box visible on the help-center home
Bug: ticket reply fails when the attachment is larger than 10 MB
Recommendation: explain the empty state on the Projects page
Sessions
and sandboxes
An accepted recommendation starts a run in an isolated sandbox with your real toolchain. Watch the agent work, send it files, steer it mid-run, or let the foreman in Slack keep you posted. Claude Code, Codex or Cursor under the hood; Starko runs the loop.
Learn more about sessions and sandboxesTemplate picker crashes when a template has no body
- searching
whatsapp-template - reading
components/functional/whatsapp-template-picker.tsx - reading
actions/connections/whatsapp.ts - running
npm test -- whatsapp-template
TemplatePicker reads template.body.length before checking the body exists; a template with no body throws on render. I also found 41 stale templates in the picker and no way to delete them, which is what Ana asked for. I will fix the crash first and add a delete action in the same pull request unless you want them separate.- editing
actions/connections/whatsapp.ts - writing
components/functional/whatsapp-template-manager.tsx - writing
tests/whatsapp-template-manager.test.tsx - running
npm run lint && npm run typecheck && npm test
git push origin starko/fix-d2728b0cReview
and merge
Every pull request opens inside Starko with its own proof: what was reproduced, every command the sandbox ran with its exit code, the reviewer agent's risk verdict, and the checks. Merge from the same panel; the ticket closes and the customer is told.
Learn more about review and mergefeat(whatsapp): add the ability to delete a WhatsApp template
starko-ai[bot]·updated 20 hours ago· Open
mainstarko/fix-d2728b0c-ad880f 9 files+1,354 -62Run
Open ticketReproduction
Reproduced
Picker listed 41 stale templates; no delete affordance anywhere in the UI.
Verification
13 commands run in the sandboxAgent’s claim: deletion removes the template on Meta and the local mirror row; the picker drops it from every send and broadcast flow.
- exit 0
npm run lint✔ 0 problems
- exit 0
npm run typechecktsc --noEmit · 0 errors
- exit 0
npm test -- whatsapp-templateTests: 8 passed, 8 total
- exit 0
npm run test:e2e -- template-manager3 passed · deleted a template in the seeded sandbox
Reviewer agent
Risk: mediumPlanning
and monitoring
Projects carry their tickets on a timeline coloured by delivery state: planned, running, pull request open, merged. Templates seed a project with its ticket distribution, and the ledger below tells you what every run cost.
Learn more about planning and monitoringStarko AI
Every fix run, what came of it, and what it cost: the ledger behind the @starko agent.
Merged PRs
41
Merge rate
85%
merged ÷ runs that opened a PR
Time to merge
3h 40m
median, run start → merge
Billed / merged PR
$18.20
platform cost $6.90
Outcomes
62 runs · 55 opened a PR · 9.4M tokens · $745 billed at list price · $283 platform cost| Ticket | Task | Outcome | Started |
|---|---|---|---|
| Delete a WhatsApp template | Feature · medium | Running | 6 min ago |
| Ticket reply fails above 10 MB | Bug · small | PR open | 1 h ago |
| Explain the empty Projects state | Feature · small | Merged | yesterday |
| Invoice PDF missing VAT line | Bug · small | Merged | yesterday |
| Annual plan proration | Feature · large | Needs info | 2 d ago |
Runs, outcomes and cost Every run is accounted for.
An accepted recommendation becomes a run: reproduce the problem, implement the change, let the reviewer agent check it, open a pull request. A revise verdict or red CI re-dispatches the run within a cap you set. The Starko AI ledger shows each run's trigger, outcome, CI state and cost, next to merge rate and time to merge.
- Runs in parallel across many tickets, not one sprint at a time.
- Per-project autonomy policy: auto-dispatch small fixes, ask on large ones.
- Merge rate, time to merge and cost per merged PR, always visible.
Bring your own coding agent Starko operates the agent you already use.
Starko is not a model. It is the system around the model: it decides what to build, writes the brief, provisions the sandbox, reviews the diff, and tracks what happened after the merge.
- Claude CodeAnthropic
- CodexOpenAI
- CursorCursor agents
- GitHubPRs, checks, merge
Self-hosted models and bare-metal runners available on Enterprise.
- of Starko pull requests get merged
- 85%
- software teams shipping with Starko
- 18
- channels feeding one factory
- 11
- charged for anything that doesn't merge
- $0
Pricing The unit of billing is a merged pull request.
Collecting feedback, AI answers and recommendations are free. Each run has a credit cap you see before you accept it. A pull request that is closed without merging costs nothing.
Free
Collect feedback, answer questions from your knowledge base, and see the recommendations Starko would build.
Start for free- All channels and the website widget
- AI answers from your knowledge base
- Recommendations with size and credit cap
- Up to 2 seats
Factory
Most teamsAccept recommendations and let runs build, review and open pull requests. Billed per merged PR.
Get a demo- Everything in Free
- Runs in isolated sandboxes with a credit cap
- Reviewer agent, evidence and in-app merge
- Automatic revise on review or red CI
- Unlimited seats
Enterprise
Self-hosted models, dedicated runners and a partner program for your customers.
Talk to us- Everything in Factory
- Self-hosted models and bare-metal runners
- SSO, audit log, data residency
- Partner and reseller program
- Dedicated support
Why a factory
Coding agents made writing code cheap. Deciding what to write, proving it works, and closing the loop with the customer who asked did not get cheaper. That is the part Starko automates.
Starko started as a support inbox for software companies with thousands of business customers. The requests those customers made were the roadmap; the factory is what happens when the inbox can act on them.
Three assumptions behind the product
FAQ Questions teams ask before they switch it on.
Connect a channel and a repository. Turn the support queue into a build queue.
Feedback starts flowing in minutes. The first recommendation shows up as soon as a pattern repeats, and the first pull request as soon as you accept one.