Launching today
TraceX
The AI engineer for your production issues
42 followers
The AI engineer for your production issues
42 followers
TraceX reads a customer's email, Slack message or Jira ticket, investigates your database, code and logs read-only, and returns the root cause with evidence for every claim, a fix as a pull request and a drafted reply. Free in private beta.
Free
Launch Team
Dropbox SignSimple, secure eSignatures for the way your team works.
Promoted
Hey Product Hunt 👋
Every engineering team knows this message: "My refund never arrived." Someone stops what they're doing, opens the database, greps the code, reads last week's deploys, checks Sentry, and an afternoon later finds that a migration renamed a state the refund worker still filters on.
We built TraceX to do that investigation for you.
How it works: an issue comes in by email, Slack, Jira or a question in the dashboard. TraceX plans its checks and works through your database, code and logs, the way your best engineer would. You watch every step as it happens. It comes back with:
• the root cause, with its confidence and the evidence for every claim, and the hypotheses it ruled out • the fix as exact code edits, opened as a pull request in one click • a reply to the customer, drafted with the facts
What we cared about most:
• Read-only. Database connections use read-only credentials, write queries are refused before they run, and code changes only ever arrive as pull requests you review. • Nothing kept. TraceX reads your systems live and keeps a note of what it looked up, not your rows. It never trains on your data. • Answers you can check. A second model checks every answer against the evidence before you see it.
It also works as a chat for your whole team ("how many refunds are stuck, and since when?"), reviews pull requests with your codebase as context, and has a CLI that answers in your terminal and works as a coding agent that asks before every change.
It's free while we're in private beta. You can try the sandbox without signing up: https://app.tracexpro.com/sandbox
I'd love your honest feedback, especially on where TraceX gets it wrong. I'll be here all day answering questions.
read-only creds is good but on prod it's not only about writes. we had an incident where our pool was saturated, the last thing I want then is an agent sending more queries on the primary. can I point it to a replica or a Neon branch?
@raphael_millet2
Totally fair. Yes, you can point it at either: TraceX just takes a connection string, so it never needs to touch the primary.
For Neon, use a read replica for live data. A branch is a snapshot from when you created it.
Load-wise, TraceX opens one connection per investigation and closes it when done. Queries are read-only, with a 3s timeout and a 20-row cap. For a hard limit on your side:
ALTER ROLE tracex CONNECTION LIMIT 2;
Lets apply for early-access: https://app.tracexpro.com/early-access
The evidence-for-every-claim part is what would sell this to me. A root-cause answer I can't check is just another confident guess, and the ruled-out hypotheses are the bit I'd actually want in front of a customer — it shows the work instead of just the verdict. Curious how it behaves when the evidence points two ways, or when the fix it proposes touches a migration — does it still draft the PR, or hold back and ask?
@jcbuilds
That's exactly the bar we hold ourselves to: an answer you can't check is just a confident guess.
Every claim links to its evidence: the exact record, log line, or code it came from. A cause with no direct evidence is capped at 30% confidence, so it can't sound sure.
On evidence pointing two ways: today TraceX scores the leading cause by the evidence for it, and lists what it couldn't settle under "Missing information". It doesn't yet put the competing explanation next to it, or show why each alternative was ruled out, as fully as you're describing. You've convinced me that's the part that matters most, especially for what you show a customer, so it's what we're building next.
On migrations: it never opens a PR on its own. It can draft the fix, but a person opens the PR, and anything touching migrations or database structure is flagged high-risk with the reason. Next we're making those open only as draft PRs, and only with an admin's approval.