“Renamed title to name in the challenge form. Pushed.”
Not a code reviewer. An integration guardrail.
No more
integration guy.
Your AI reviewer reads the diff. Meld checks the whole app: it maps every call your frontend makes to your backend, flags the ones a pull request breaks, then starts your stack and replays them to prove it.
Meld checkPR #42 · 1 break found
FFrontendsends { name }
BBackendexpects { title }
RRunning stack422 · title missing
Pull request check just now
ROUTES THAT MATCH✳PAYLOADS THAT FIT✳STACKS THAT START✳ROUTES THAT MATCH✳
01The old way
Everyone’s code works.
Together, it doesn’t.
AI helps every teammate ship more code than ever, and every pull request passes review on its own. But nobody holds the whole picture. The frontend renames a field, the backend moves a route, a new setting never reaches the server, and you find out at the demo. Usually one person ends up stitching it all together at 3 a.m.
Moved /api/universities to /api/institutions. Didn’t touch the UI.
“Submit does nothing and the list is empty. Who changed what?”
02How Meld makes your life easier
Less stitching.
More shipping.
Checking that every piece still fits is the part of integration nobody wants. Meld does it on every pull request, so nobody has to.
Find out at the pull request, not the demo
Without MeldSubmit does nothing, the list is empty, and nobody knows who changed what.
With MeldMeld flags the pull request that broke the app before you merge it, and names the call it broke.
Nothing to set up
Without MeldWriting contract tests and API specs nobody has time for.
With MeldInstall the GitHub App and pick a repo. Meld follows it on its own, with no specs, tests or config to write.
Know exactly what to fix
Without MeldAn hour of console.log to work out which side changed.
With MeldEach break comes with the file, the line and the failing request, explained in plain language.
Fix it in one paste
Without MeldExplaining the bug to your coding agent from scratch.
With MeldCopy the handover prompt into Cursor or Claude Code. It’s built from Meld’s findings, so the agent starts with the facts.
Works alongside your AI reviewer and CI. Nothing to replace.
03The Meld flow
Every PR.
Checked end to end.
Meld follows your repository on its own, reads what each change touches, and flags only what that change broke.
01
contract checked ✓
Read every change
Meld maps routes, frontend calls, request bodies, imports, function signatures and config, then compares each pull request with the code it changes.
02
F
Frontend sends nameSubmitChallenge.jsx:8
B
Backend needs titleChallengeCreate model
M
Meld replayed it422 · title missing
Prove it on the running stack
On each pull request a GitHub Action starts your docker compose stack, signs in, and replays your frontend’s real requests against the backend.
03
Fix it fast
Every finding comes with the file, line and evidence, explained in plain language, plus a prompt you can paste straight into your coding agent.
04What Meld believes
Facts first.
Then words.
01
Evidence over opinion.
Every finding comes from a deterministic check with a file and line. The AI explains findings; it can’t invent or change one.
02
Only what this change broke.
Meld compares each change with the code before it, so old issues don’t bury the one you just introduced.
03
Your code stays yours.
Your stack runs on GitHub’s runners, not on Meld. Meld only sends the paths to request, never code, and you sign in with GitHub. What Meld can access ↓
05What Meld can access
Read-only.
Easy to undo.
Meld needs to read your code to check it. Here is exactly what it can reach, where your code goes, and how to take access back.
Reads code. Writes only a check.
The Meld GitHub App reads code, pull requests and repository metadata, and posts a “Meld” check with its findings on your pull requests. It can’t push, comment, merge or change settings. Signing in asks GitHub only for your name and email.
Only the repos you pick.
You choose repositories in GitHub’s own picker when you install the App. Meld can’t see the rest, and it stops following a repo as soon as you deselect it.
Short-lived, never saved.
Each time Meld needs GitHub, it asks for a token that expires within an hour. Tokens aren’t stored in the database or your browser, and there are no passwords.
No code goes to an AI company.
Meld keeps a copy of each connected repo on its own server to analyse it. The AI model that explains findings runs on that same server, so nothing is sent to OpenAI, Google or anyone else. The runtime check runs on GitHub’s runners and reports only its results.
One click to remove.
Remove a repo in Meld to delete its copy and every analysis. To cut access completely, uninstall the App in GitHub → Settings → Applications; Meld can’t read anything after that.
Open to everyone
Ship together.
Demo calmly.
Anyone can try Meld. Sign in with GitHub, install the Meld app on a repository, connect it in Meld, and your next pull request gets a check.