Troubleshooting Business Knowledge

Private beta

Business Knowledge is in private beta. If Business Knowledge isn't in your PostHog sidebar, your project isn't in the beta yet. Send feedback in the app.

Fixes for the problems that come up most often. Each fix links to the page with the full details.

An agent can't find a fact you added

  1. Open the source and check its status. Search reads a source that shows ready or indexing. An indexing source is found by keyword only, until its semantic index is built.
  2. Confirm your organization approved AI data processing. Without it, the content check doesn't run and search leaves every document out.
  3. A new or changed document waits for the content check, which runs about once an hour.
  4. Ask the question in the Playground Search tab. If the chunk appears there and your own agent doesn't find it, check which library the agent's key covers.
  5. If the wording of the question is far from the wording of the source, wait for semantic search. The source shows indexing until it's ready.

A source stays on indexing

Indexing means keyword search works and the semantic index is still building. It can take up to an hour. Large sources take more than one pass.

If the source shows indexing for several hours, check that AI data processing is still approved for the organization.

A URL source fails

MessageWhat to do
URL is not reachable. or URL is not reachable from this environment.PostHog can't reach the address. The page has to be public. Private and internal addresses are refused.
Remote responded with status ...The site returned an error. Open the URL in a private browser window. A page behind a login can't be indexed.
Could not extract any text from the URL.The page has no readable text, or it builds its content with JavaScript after it loads. Add the text as a text source.
Sitemap is not valid XML.The sitemap URL doesn't return a sitemap. Use the Same-origin crawl option.
Remote response exceeds the maximum allowed size.The page is over 10 MB.
Another source is already being processed. Please wait and try again.A project processes one URL source at a time. Add the next one when the first is ready.
Processing timed out. You can retry via refresh.Click Refresh on the source.

A crawl skips pages that robots.txt disallows for the PostHog-BusinessKnowledge crawler. If pages are missing, check your robots.txt, the Max pages and Max depth settings, and Skip paths.

When a refresh fails, the row shows refresh failed and the previous text keeps answering questions.

A file doesn't upload

MessageWhat to do
Unsupported file type.Upload a PDF, DOCX, Markdown, CSV, or plain text file
File exceeds the 50 MB cap.Split the file
PDF contains no extractable text (it may be scanned/image-only).The PDF holds images of pages. Run text recognition on it first, or paste the text.
Encrypted PDFs are not supported.Remove the password and upload again
PDF has ... pages (cap is 2,000).Split the PDF
CSV has no header row.Add a first row with column names

A CSV file keeps its first 10,000 rows.

A source shows unsafe content

The content check flagged at least one page, and search leaves that page out. On a URL source, Indexed pages shows which page. Remove the text that reads as an instruction to an AI agent, then refresh the source.

You reach a limit

Knowledge source quota exceeded for this project. means the project is at a limit for sources or chunks. Delete sources you don't need, or lower Max pages on a large URL source.

The Playground refuses questions

The Playground needs AI data processing approved for the organization. An organization owner or admin turns it on.

Your own agent gets an error

See troubleshooting for connected agents.

A repository search finds nothing

Repository search matches file names, path fragments, and identifiers. It doesn't search inside files. Pass billing-policy or README, not a whole sentence. See GitHub repositories.

Still stuck

Send feedback in the app and include the name of the source.

Still have questions?

Was this page useful?