How TraceX handles your data
What TraceX reads from your systems, how queries are checked, what reaches AI models, how secrets and personal data are masked, how credentials are stored and what TraceX keeps.
Connecting a tool to production is a real decision, and you should know exactly what it can do. This page explains what TraceX reads, what it sends to AI models, what it stores, and the safeguards around each.
In short: TraceX reads your systems live and read-only, only while working on an investigation or a question. It masks secrets before anything reaches an AI model, records every read, and keeps no copy of your databases or repositories.
What TraceX reads
Section titled “What TraceX reads”TraceX reads your systems live, read-only, only while working on an investigation or a question:
- Databases: every query is parsed and refused unless it is a single read-only
SELECT(no writes, schema changes, locks, file or remote functions), and runs in a read-only transaction with row and time limits. BigQuery scans are capped at 20 GB a question. - Code: read through the GitHub App or your read-only token; not copied.
- Logs and errors: searched with your read-only keys around the time of an issue.
It keeps no copy of your databases or repositories.
What reaches AI models
Section titled “What reaches AI models”To investigate, TraceX sends AI models the text they need: the issue, relevant query results, log lines and code. Before anything is sent:
- Secrets, keys and tokens are always masked. This can’t be turned off.
- With hide personal data on, emails, phone numbers, card numbers, national ids, IBANs and IP addresses are replaced with placeholders too, and put back in what your team sees.
On Enterprise you can use your own AI provider keys.
AI safeguards
Section titled “AI safeguards”Content TraceX reads can contain text written by anyone, such as a customer email or a ticket comment, so it is treated as data, never as instructions. Every question and every piece of content TraceX reads is checked for prompt extraction, jailbreaks and hidden instructions, and answers are checked before they are shown. What the safeguards stopped is listed under Security → AI safeguards.
Credentials
Section titled “Credentials”Connection passwords, keys and tokens are encrypted at rest and never shown again after you save them. Every credential is checked with a real read when you add it, so a wrong or over-privileged key is caught early. GitHub uses short-lived installation tokens, not stored personal tokens.
Every read is logged
Section titled “Every read is logged”Every query, log search and file read is recorded in the data access log, with the investigation or question it was for. Export it as CSV for your own review.
What TraceX keeps
Section titled “What TraceX keeps”- investigations and their reports,
- conversations,
- the evidence summaries they cite,
- usage records,
for as long as your retention settings say. You can export everything first. Shared links never include raw data (Share a report).