Add and manage sources

Open Business Knowledge and choose Add source.

Text

Paste a policy, an FAQ, or any other text, and give the source a name. Edit the source when the text changes.

Text is capped at 1 MB. Split a longer document into more than one source.

Saving rewrites the chunks for that source. An agent that already has the old text in its conversation keeps it until that conversation is refreshed.

URL

Paste a public http or https URL and choose how PostHog should read it. PostHog rejects a URL it can't reach, including a private or internal address, with URL is not reachable from this environment.

• Single page fetches that page.
• Sitemap.xml reads the sitemap at that URL, or /sitemap.xml on the same site, and indexes each listed page.
• Same-origin crawl indexes that page and other pages on the same site, staying under the page's path. A URL at the site root covers the whole site. Skip paths you don't want. Depth and max pages bound the crawl. The crawl follows robots.txt. The crawler identifies itself as PostHog-BusinessKnowledge.

Skip paths are URL path patterns, matched with fnmatch. Put one pattern per line, or separate them with commas. /docs/private/* skips that section.

Max pages defaults to 200. The cap is 5,000. Max depth applies to a same-origin crawl. It defaults to 2 and can't be higher than 5.

Include globs, under Advanced: override include scope, replace the path PostHog derived from the entry URL. Leave them empty to keep that path.

PostHog doesn't index a page body larger than 10 MB. A failed refresh keeps the previous text, and the row shows refresh failed. The old chunks still answer questions.

File

Upload a PDF, DOCX, Markdown, CSV, or plain text file. The cap is 50 MB. PostHog rejects a PDF with more than 2,000 pages. PostHog truncates a CSV at 10,000 rows.

PostHog stores the text it reads from the file. It doesn't keep the original file. To change a file, delete the source and upload it again. You can still rename the source and change Always include in replies.

Auto-refresh

A URL source can refresh on a schedule: every hour, every 6 hours, every day, or every week. Manual only means PostHog fetches the source when you add it, and again when you refresh that row yourself.

Changing the auto-refresh interval doesn't fetch the source immediately. Changing the URL or the crawl settings does.

Text and file sources don't refresh on a schedule.

Always include in replies

Turn this on for guidance that should be in every reply: tone, policies, and company direction. PostHog still runs the content check. PostHog includes the source even when the question doesn't match it.

The text injected from these sources is capped at 20,000 characters combined. PostHog leaves out content past that cap.

Source status

If the organization hasn't approved AI data processing, a ready source can still show that semantic search is off. Search leaves out documents that haven't passed the content check, for both keyword and semantic search.

Content checks

PostHog checks each document before using it in an answer. Search leaves out a document marked unsafe. Search also leaves a document out until that check finishes. The check doesn't run until the organization has approved AI data processing.

On a URL source, Indexed pages lists each page as safe, unsafe, or pending. Pending means search skips that page until the check finishes. Unsafe means the page stays listed and search excludes it.

The source shows unsafe content when it has at least one flagged page.

Edit or delete a source

Open a row to change it.

• A text source: edit the name and the content.
• A URL source: edit the name, URL, crawl settings, auto-refresh, and Always include in replies. Use Refresh to fetch it now.
• A file source: edit the name and Always include in replies. Replace the file by deleting the source and uploading again.
• A learned source: edit or delete it. See Learn from support.

Delete removes the source and its chunks. PostHog records the change in the activity log.

Limits