Skip to content
TraceXDocs
Ask TraceX

Knowledge base

Give TraceX your runbooks, architecture docs and postmortems so it knows your system the way your team does. Upload, paste, or sync from Notion and Confluence.

Your databases and code say what your system does. They don’t say why it was built that way, which alerts are noisy, what the runbook for a stuck queue is, or what went wrong last time. That knowledge lives in docs and in people’s heads.

The Knowledge base is where you give it to TraceX. It searches a project’s knowledge base while it investigates and answers, so it knows your system the way your team does: what services exist, known issues, runbooks and past incidents.

What to add

Section titled “What to add”

The documents that help most are the ones you’d hand a new on-call engineer:

  • Runbooks: what to do when a queue backs up, a job fails, a provider is down.
  • Architecture notes: which services exist, what talks to what, where data lives.
  • Postmortems: past incidents, their causes and fixes. They help TraceX recognise a repeat.
  • Glossaries: your product’s terms and what they mean in the data.
  • Known issues: quirks and workarounds your team already knows about.

Add documents

Section titled “Add documents”

On Knowledge base:

  • Upload Markdown, text, reStructuredText, HTML, PDF or Word (.docx) documents.
  • Paste a document. Markdown headings (#, ##) split it into sections TraceX can find, so give each topic its own heading.
  • Connect Notion or Confluence to keep pages in sync (Notion and Confluence).
  • For Google Docs, download the doc as .docx or PDF and upload it.

Check what TraceX would read

Section titled “Check what TraceX would read”

Use the search box on the Knowledge base page to see which sections TraceX would read for a question. If the right section doesn’t come up, add a heading or a sentence with the words people actually use.

Search by meaning

Section titled “Search by meaning”

When search by meaning is turned on for your organization, TraceX also finds sections worded differently from the question, for example “payment stuck” finding a runbook titled “Capture job failures”. Document text is sent to the embedding provider (with secrets masked first), like every AI call.

Section titled “Related”