Knowledge base — documents (power user).

If you have existing content (a help-centre URL, a product manual PDF, a Word policy doc), you can ingest it directly instead of typing FAQs by hand.

Supported formats

How ingestion works

Knowledge Base tab → Add a source. Either paste a URL or pick a file. Click Ingest.

The dashboard shows the document immediately in the Imported documents list with status pending. Behind the scenes:

1. Fetch — Tenlo downloads the URL or reads the uploaded file
2. Parse — extract clean text (URL: Readability; PDF: text extraction; Word: raw text; MD: as-is)
3. Chunk — split into ~500-token chunks with 50-token overlap so the bot can match precisely without hallucinating across boundaries
4. Embed — generate a vector for each chunk
5. Index — save to your private search index

Status pill flips through pending → processing → embedded (success) or failed (with an error message). Typical end-to-end time: 30 seconds to 2 minutes per document. Scanned PDFs show a Processing (OCR)… badge while Mistral OCR runs (usually 1–3 minutes for a 10-page scan).

Limits

• 100 documents per business (lifetime) — shared across Support Chatbot (P01) and Sales Assistant (P05). Deleting a document anywhere on either product frees a slot for either product.
• 25 MB max per file
• 20 ingest operations per hour — a sitemap crawl that fans out into 80 pages still counts as 1 operation against this cap. Single URL or single file = 1 operation.
• Scanned PDFs — first 50 pages are OCR’d; beyond that, processing stops and the document card shows how many pages were skipped. Re-upload a split version if you need the tail.
• Sitemap crawl — same-host pages only. CDN subdomains or off-host URLs found in the sitemap are filtered out. Already-ingested URLs are detected and shown as such (no double-ingest).

Deleting an ingested document

The Imported documents list has a Delete button per row. Clicking it:

• Removes all the document’s chunks from the search index
• Deletes the file from storage (if uploaded)
• Removes the row from the list

The deletion is clean — the bot stops citing that document immediately.

FAQs vs. documents — when to use each

You can mix them freely. The bot searches both at once and returns the best match, regardless of source type.

What’s not yet supported

• JavaScript-rendered pages (single-page apps that need JS execution to show content)
• Scheduled re-ingest — today, you re-ingest manually after content changes
• Multi-file batch upload — sitemap crawl covers the URL side; file batches are still one-at-a-time
• OCR for scanned PDFs beyond 50 pages — anything over the cap is truncated with a notice on the doc card

These are on the roadmap. For now, the workaround is to ingest URLs/files individually.

Sitemap crawl (V2.1)

Shipped 2026-05. Lets you bring in a whole help centre or docs site in one go without pasting URLs one at a time.

How it works

1. Knowledge Base tab → Sitemap crawl card. Paste any URL on the target site (e.g. https://docs.example.com/getting-started).
2. Tenlo probes /sitemap.xml and /robots.txt for a sitemap and parses it (recursing one level into sitemap-index files).
3. You get a preview list of every URL the sitemap exposes. Pages under the same section as the seed URL are preselected; URLs you’ve already ingested are flagged and disabled.
4. Tick the pages you want, click Ingest selected. Each URL becomes its own kb_documents row and runs through the same fetch / parse / chunk / embed pipeline as a single URL.

Constraints

• Same host only — sitemap entries on different hosts (CDN subdomains, partner domains) are filtered out
• Slot accounting — if a crawl would push you past the 100-doc cap, the UI tells you how many slots remain and caps the selection
• One operation against the rate limit regardless of how many pages were enqueued
• Atomic enqueue — if the queue rejects the batch, all just-inserted rows are rolled back so they don’t ghost-occupy the cap
• No sitemap found? Tenlo can’t discover one ~30% of the time (especially on hand-rolled marketing sites). Fall back to single-URL ingestion.

Common patterns

• Help-centre import — paste your /help landing page → ingest the section in one click
• Product-docs migration — paste the docs root → tick the categories that map to support-ish questions, ignore developer-only pages
• Refresh — re-run the crawl periodically; already-ingested URLs show as such and stay unticked by default