Ingestion

Ingestion reads your product's content and turns it into the knowledge the AI agent uses to demo, onboard, and support users.

After you've created a site, ingestion is how you feed Floe the information it needs. Reading, processing, and indexing happen automatically.

Ingestion is for material that already lives in a supported source. For a short authoritative fact that is missing from those sources, an Owner or Admin can write a Reference Guide directly under Knowledge → Facts. A Reference Guide supplies facts for answers; it does not become a walkthrough.

How ingestion works

You give Floe the URLs to your content. What Floe reads there becomes the knowledge the agent draws on during a demo call or when helping a user with in-app guidance.

Supported content types

• Documentation: Product docs sites. Best for feature references, setup guides, API docs.
• Knowledge Base: Help center articles. Best for FAQ, troubleshooting, how-to guides.
• Changelog: Release notes. Best for what's new, version history.
• YouTube Video: Video transcripts. Best for product walkthroughs, tutorials.
• Loom Video: Loom recordings. Best for internal demos, feature explanations.
• Image Gallery: Screenshots, diagrams. Best for visual product reference.

Adding an ingestion config

When you create a site in the dashboard, Floe adds the website at its product URL as the first source, listed as Your website, and starts reading up to 50 of its pages. Re-running it reads up to 50 pages again. Add your other content as sources of its own:

1. Go to your site in the dashboard
2. Open Knowledge → Sources
3. Click Add a source (or Add source once you have one)
4. Select the content type
5. Enter the source URL. For a docs site, help center, or changelog, Floe follows links from that page to the other pages on the same site, so start from the main page. Each run of a source reads up to 20,000 pages.
6. Choose Add and run now to crawl just that source straight away, or Add without running to crawl it later

You can add multiple sources per site. A typical setup might include your docs site (Documentation), your help center (Knowledge Base), and a few product walkthrough videos (YouTube). Run all sources crawls every source again; Re-run on a source crawls only that one.

To change a source's address, use Edit on it. Editing doesn't crawl; re-run the source to read it from the new address.

Uploading documents

For facts that don't live on a crawlable page, such as spec sheets, one-pagers, or sales collateral, upload PDFs under Uploaded documents on Knowledge → Sources. Each file can be up to 25 MB and must contain text; scanned or image-only PDFs are not supported.

Your agent can answer from an upload as soon as it finishes. A new upload also refreshes your product knowledge within a few minutes, and several uploads in a row share one refresh. If that refresh can't be scheduled, the upload message says so; use Re-run research on Knowledge → What Floe learned to include the file. Uploading the same file again changes nothing. Your uploaded documents are listed on the tab; remove one from Knowledge → Library.

Content handling

Floe handles JavaScript-heavy sites and single-page applications. It stays on the site the source's address is on.

Running ingestion again won't create duplicate entries. A run reads every page in the source again: edited pages are updated, new pages are added, and unchanged pages are left as they are. See Keeping content up to date.

Content access and retention

Self-service ingestion currently accepts publicly accessible URLs. Sources behind a login are not supported. Do not include credentials, cookies, or access tokens in a source URL.

Floe processes the public documentation and other content you configure as Customer Data only to provide the service. Customer Data is retained for your subscription and deleted within 30 days after termination, unless law requires otherwise; backups are purged within 30 days of deletion. Access is limited to your organization, authorized Floe personnel, and the service providers needed to operate Floe, as described in the Privacy Policy. Removing a source does not delete already-ingested content; remove that content separately from Knowledge → Library. This policy is separate from session-recording retention.

Monitoring ingestion

Knowledge → Sources shows each source's status, how many pages Floe has from it, and when it was last crawled:

• Queued: waiting for its turn in a run
• Crawling: being read now
• Ready: its last run finished
• Failed: its last run produced nothing; the reason is shown under the source
• Not crawled yet: added but never run

A run in progress can be cancelled. Library lists every page read, with the source it came from.

Typical processing time is 2-5 minutes for a small docs site (50-100 pages).

Keeping content up to date

After a run, Knowledge → Sources shows how many pages were updated, how many are new, how many came back, and how many are gone.

When Floe can no longer reach a page, because it returns "not found", redirects somewhere else, or stays unreachable across runs, the page is marked Gone in Library, and your agents stop answering from it. A later run that reaches the page again brings it back. A run that can't read a source in full, for example during an outage, marks no page gone. When a run finds changes, Floe also updates what it learned from your content (see Research).

Refreshing on a schedule

Turn on Automatic refresh on Knowledge → Sources and choose Daily, Weekly, or Monthly to run all your sources on that schedule. It needs at least one website source: documentation, a knowledge base, a changelog, or a custom URL. Video and image sources on their own are not refreshed on a schedule. It is off until you turn it on, and it is set per site. A scheduled run starts some time after it is due, not at an exact time.

What's next

Once content is ingested, Floe prepares the capabilities and guided workflows the agents use. See the Quickstart for the full flow, or learn more about how it works.

Reference Guide edits do not require another ingestion run; a guide is available to the agents as soon as it is Ready.

FAQ

Can I ingest content behind a login? No. Ingestion currently accepts publicly accessible URLs only. Sources behind a login are not supported, and you must not provide credentials, cookies, or tokens in an ingestion configuration.

What if my docs site changes frequently? Turn on Automatic refresh, or select Run all sources after a change: each run picks up edited and new pages. The generated demo context is refreshed by Floe on request, so tell us after a significant change.

Is there a limit on how much content I can ingest? Each run of a source reads up to 20,000 pages. That limit is per run of one source; there is no limit on how many sources a site can have. Most products ingest 50-200 pages. Larger knowledge bases work fine; they just take longer to process.