An issue comes in
A support email, a Zendesk or Intercom ticket, a PagerDuty alert, a Slack thread, a Jira or Linear issue, or a question in the dashboard.
Every customer issue, traced to the root cause.
TraceX is the AI engineer for your production issues. It reads the support email, investigates your database, logs and code, finds the root cause with evidence, and writes the fix.
M
Maria Chen · to support · 9:41
My refund from last week still hasn't arrived. Order #48213.
Intake
Triage
Evidence
Root cause
Action
Resolve
Reading the email…Checking database…Reading the code…Checking Sentry…Root cause found
PostgreSQL 34 refunds with state = 'pending' since Oct 21
GitHub Migration 0142 renamed pending → awaiting_payout
Sentry No errors: the worker finds nothing to process
Live traceread-only
Read the emailorder #48213 · refund
Planned 4 checksdata · code · errors · logs
SELECT state FROM refunds WHERE order_id = 48213'pending'Searched code for the refund filterrefunds.py:42
Read 3 recent commits7f3b10 renamed the state
Checked Sentry · refund-worker0 errors
Tested: the worker skips the renamed stateconfirmed
Root cause refund worker
Migration 0142 renamed the refund state,but refunds.py still filters on pending.
Confidence93%
ExpectedRefunds move to the payout queue within an hour.
ActualThe worker filters on
pending, which no longer exists.Reply drafted
Hi Maria, we found the problem and your refund is on its way.
Connects read-only to the tools your team already runs.
PostgreSQLSee it work 
MongoDBSee it work 
BigQuerySee it work 
SnowflakeSee it work GitHubSee it work 
GitLabSee it work 
BitbucketSee it work SentrySee it work 
DatadogSee it work 
CloudWatchSee it work 
Grafana LokiSee it work 
ElasticsearchSee it work 
OpenSearchSee it work 
Google Cloud LoggingSee it work 
Azure MonitorSee it work Gmail inboxSee it work 
SlackSee it work 
JiraSee it work 
LinearSee it work 
ZendeskSee it work 
IntercomSee it work 
PagerDutySee it work 
OpsgenieSee it work 
Notion
ConfluenceNow in private beta with early engineering teams, free while in beta.
Integrations
25
databases, code, logs, support desks and on-call tools
Visible steps
6
in every investigation, from intake to resolve
Production writes
0
read-only access and read-only queries, always
Free in beta
50
investigations included during the private beta
How it worksFrom a customer email
See an investigation From a customer email
to a pull request.
TraceX investigates
It queries your database or warehouse read-only, reads the code and recent changes, checks your logs and errors, and reads your runbooks.
It proves the cause
A root cause with its confidence, what should happen versus what does, and the evidence for each claim.
It writes the fix
A pull request with the change and how to validate it, and a reply ready for your customer.
Investigations
The root cause, with the evidence to prove it.
TraceX works the issue the way your best engineer would: it reads the thread, plans what to check, and follows the data and the code until the cause is clear.
- Root cause, confidence and expected versus actual behaviour
- Every finding cites the record, code or log line behind it
- A fix as a pull request, and a drafted reply to the customer
PostgreSQL 34 refunds with state = 'pending' since Oct 21
GitHub Migration 0142 renamed pending → awaiting_payout
Sentry No errors: the worker finds nothing to process
Root cause refund worker
Migration 0142 renamed the refund state,but refunds.py still filters on pending.
Confidence93%
ExpectedRefunds move to the payout queue within an hour.
ActualThe worker filters on
pending, which no longer exists.Conversations
Ask about your systems in plain English.
How many refunds are stuck? Which customers were charged twice? What changed in checkout since Monday? TraceX looks it up and shows its work.
- Answers from your databases, code and logs, never from guesswork
- Each step shown as it happens, each claim cited
- Spreadsheets, CSV and PDF exports when you need them
How many refunds are stuck, and since when?
Queried refunds · 34 rows
Read 3 commits in payments-api
Checked the answer against the evidence
34 refunds are stuck in
pending, all created after migration 0142 shipped on Oct 21 1 2.
| Day | Stuck |
|---|---|
| Oct 21 | 12 |
| Oct 22 | 14 |
| Oct 23 | 8 |
Fixes and code review
From the root cause to a pull request.
TraceX writes the smallest change that fixes the cause, explains its risk and how to validate it, and reviews your team's pull requests too.
- A reviewed diff with its risk level
- Validation steps in the description
- Reviews on every PR, with one-click suggestions
app/workers/refunds.py+1 −1
def pending_refunds(db):
− return db.refunds.filter(state="pending")
+ return db.refunds.filter(state="awaiting_payout")
Validate run the refund worker against staging; 34 refunds move to payout
Learns from your team
Tell it once. It remembers.
Say whether a root cause was right, and what it really was. The next time something similar comes in, TraceX checks your answer first, flags an issue that came back after its fix, and says which missing source would settle what it couldn't confirm.
- Your corrections guide later investigations, as leads it still proves
- Repeat issues grouped, with a warning when a merged fix didn't hold
- Past incidents replayed as accuracy tests after every model change
Root cause refund worker · 62%
The refund worker's retry limit was reached, so refunds stay pending.
RightPartlyWrong
What was the actual cause?The refund.updated webhook was turned off in Stripe.
Saved · used on similar issues
Three weeks later
Refunds stuck again · 3 customers wrote in
Your team said this was the refund.updated webhook last time. Checked it first: it's off again.
Seen 2 times · back after fix #142
Connected by designYour whole stack.
Your whole stack.
Read, never written.
TraceX reads your databases, code, errors, logs and inboxes together, so every answer comes from the whole picture. Access is read-only and you can revoke it any time.
Explore integrationsIntegrations in actionWatch it work
All integrations Watch it work
with your stack.
Read-only queries, real answers
PostgreSQL, MongoDB, BigQuery and Snowflake, through read-only credentials. TraceX writes the query, runs it, and keeps only what it found.
PostgreSQL· production · read-onlySELECT id, order_id, state, created_atFROM refundsWHERE state = 'pending' AND created_at > '2026-10-21'
| id | order_id | state | created_at |
|---|---|---|---|
| rf_9921 | 48213 | pending | Oct 21 09:12 |
| rf_9934 | 48190 | pending | Oct 21 11:40 |
| rf_9958 | 48236 | pending | Oct 22 08:03 |
| rf_9987 | 48271 | pending | Oct 23 16:55 |
| … 30 more rows | |||
34 rows · read-only query · no rows stored
34 refunds are stuck in pending, every one created after Oct 21.
The change behind the problem
Code, commits and deploys through the GitHub App, or GitLab and Bitbucket with a read-only token. TraceX finds the change that caused it and writes the fix.
payments-api· recent commitsRetry card errors in checkout a1c9e2 · Oct 22
Migration 0142: rename refund state pending → awaiting_payout 7f3b10 · Oct 21
Root causeBatch EU payouts 2d8e44 · Oct 20
Fix refund worker state filter #318
OpenOpened by TraceX from the investigation · 1 file changed
12 tests passedReview requested
The migration renamed the state; refunds.py still filters on the old one.
Errors, traced to the line
Issues and stack traces, matched to the deploy that started them.
checkout-web· issuesErrorTypeError: Cannot read properties of undefined (reading 'amount')
chargeCard · first seen Oct 22, 14:02
node_modules/stripe/lib/charges.jscreatesrc/checkout/payment.ts:88chargeCardsrc/checkout/submit.ts:41onSubmitIt started with deploy a1c9e2: on a retry the payment intent is undefined.
The logs around the moment
TraceX searches the logs at the time of the issue and picks out the lines that explain it. Grafana Loki, Elasticsearch, Google Cloud and Azure logs too.
Logs· service:checkout14:01:52INFOPOST /charge 200 order=48188
14:02:03WARNcard_error retry=1 order=48190
14:02:04INFOPOST /charge 200 order=48190
14:02:04INFOPOST /charge 200 order=48190
14:02:31WARNcard_error retry=1 order=48197
14:02:32INFOPOST /charge 200 order=48197
Each card error retried the charge without an idempotency key, so 6 orders were charged twice.
Logs Insights, written for you
TraceX writes the Logs Insights query, runs it, and reads the events.
CloudWatch Logs Insights· us-east-1fields @timestamp, @message| filter @logStream like /refund-worker/| sort @timestamp desc
10:03:00INFOrefund-worker: found 0 refunds to process
10:02:00INFOrefund-worker: found 0 refunds to process
10:01:00INFOrefund-worker: found 0 refunds to process
10:00:00INFOrefund-worker: found 0 refunds to process
The worker runs every minute and finds nothing: its filter no longer matches any refund.
Starts from the support inbox
Each new customer email starts an investigation, and ends with a reply ready to send.
support@· inboxMaria Chen
9:41Refund still not received · My refund from last week still hasn't arrived. Order #48213.
Investigation started
Root cause found · 93%
Reply drafted by TraceX
Hi Maria, thanks for your patience.We found the problem and fixed it.Your refund is on its way.
Work the issue in the thread
Mention TraceX in a thread, follow up without mentioning it again, or ask it to investigate. Say whether it was right, draft the fix and open the pull request, all from the thread.
#support· SlackP
Priya 2:14 PM
@TraceX why was order 48190 charged twice?
TraceX APP
A card error retried the charge without an idempotency key, so the second attempt charged again 1. 6 orders were hit on Oct 22 2.
RightWrongDraft a fix
Tickets that come back solved
TraceX picks up the ticket, comments the root cause and links the fix. Linear issues work the same way.
SUP-212· SupportCheckout times out on Safari
To doIn reviewReported by 3 customers · Priority high
TraceX commented
Safari blocks the cookie the 3-D Secure frame needs, so the challenge never loads and checkout waits until it times out 1.
Fall back to a 3-D Secure redirect on Safari #321
Linked pull request
Starts when the alert fires
A triggered incident starts an investigation at once, and the root cause lands on the incident as a note. Opsgenie works the same way.
PagerDuty· checkout-apiP1
Checkout error rate above 5%
TriggeredAcknowledged#88 · triggered 14:04 · high urgency
TraceX is investigating
Root cause · 91%
Note from TraceX · 14:07
Deploy a1c9e2 retries card errors without an idempotency key, so retried charges go through twice. Started 14:02, 128 errors in chargeCard.
Fix: payments-api #322 · ready for review
The investigation started when the alert fired, not when someone opened a laptop.
A note on the ticket, a reply ready
Tag a ticket for TraceX: it investigates and adds an internal note with the cause and a suggested reply, never sent on its own. Intercom works the same way.
Zendesk· ticket #4512Charged twice for the annual plan
10:12Sam Park · "I upgraded yesterday and see two charges on my card."
Internal note · only your team sees it
The upgrade call timed out and was retried without an idempotency key, so the provider charged twice. 4 customers since Oct 22.
Suggested reply · not sentHi Sam, you were charged twice because of an error on our side.We've refunded the second charge today.
Support answers in minutes. Engineering gets the cause and the fix, not the complaint.
TraceX CLI
In your terminal, and in your repo.
Sign in through your browser, link a repository with one command, and ask TraceX from your terminal. In a repository that isn't linked, it works as a coding agent that asks before every change.
tracex -cfor answers with the same evidence as the dashboard- Coding mode: reads, edits and runs tests in your repo, with your approval
- Install with one line on macOS and Linux
$ tracex -c› fix the refund worker filter Edit app/workers/refunds.py - state="pending" + state="awaiting_payout" Apply? (y/a/n) y Run pytest tests/test_refunds.py? (y/a/n) y ✓ 12 passed
Built for trustPowerful by design.
Explore security Powerful by design.
Controlled by you.
Read-only by default
TraceX connects to your databases with read-only access and runs only read-only queries. Nothing in production is changed.
Evidence, not guesses
Every root cause and answer cites the records, code and log lines it came from, and a second model checks the answer against them.
Your records aren't kept
No index of your code or data. Answers keep what was looked up and where, not the rows; exported files are deleted after 24 hours.
You approve every change
Fixes arrive as pull requests for your review. In your terminal, TraceX asks before every edit and command.
Enterprise sign-in
SSO, two-factor sign-in, roles for admins, investigators and viewers, revocable sessions and a full audit log.
Your own model keys
Run TraceX on your own Anthropic or Gemini account, so AI usage is billed to you, never through TraceX.
Product tour
See TraceX in 96 seconds.
A customer email, the investigation, the root cause and the fix, start to finish. Turn the sound on.
FAQ
About TraceX
What is TraceX?
TraceX (tracexpro.com) is an AI engineer for production issues. It takes a customer report, investigates your database, logs and code with read-only access, finds the root cause with evidence, and writes the fix as a pull request, with a reply ready for your customer.
Who is TraceX for?
Software teams that lose hours tracing what went wrong for a customer: support teams handling escalations, on-call engineers working incidents, teams debugging payments and billing, and product, operations and finance teams who need answers from production data.
What does TraceX connect to?
PostgreSQL, MongoDB, BigQuery and Snowflake; GitHub, GitLab and Bitbucket; Sentry, Datadog, CloudWatch, Grafana Loki, Elasticsearch, Google Cloud Logging and Azure Monitor; Gmail, Slack, Jira, Linear, Zendesk and Intercom; PagerDuty and Opsgenie; Notion and Confluence.
Is it safe to connect TraceX to a production database?
TraceX uses read-only credentials and checks every query before it runs, refusing anything that would write. It keeps no index of your code or data, and code changes only arrive as pull requests your team reviews.
How is TraceX different from Sentry or Datadog?
Sentry and Datadog show errors and logs. TraceX reads them together with your database and code, starting from the customer's report, then gives you the root cause, the evidence for each claim, a fix as a pull request and a drafted customer reply.
How much does TraceX cost?
TraceX is free during the private beta: 50 investigations in total, 500 chat messages a month and up to 5 seats. Larger teams can book a call about higher limits, private tenancy or self-hosting.
Is this related to other products called TraceX?
No. This TraceX is the AI engineer for production issues at tracexpro.com. It isn't connected to other products with the same name, such as embedded-system trace tools or supply-chain traceability software.
Find the root cause.
Ship the fix.
TraceX is free during the private beta. Connect a database and a repository and ask your first question in minutes.
