Knowledge Base

The Knowledge Base is where you store everything your Finn needs to know about your business. Menu, pricing, hours, policies, FAQs — anything callers might ask about. When the caller asks a question, the Finn pulls the answer from here.

Don't overthink it. Most users start by uploading one PDF (their FAQ or a product brochure) and that's enough for the Finn to handle 80% of caller questions. You can always add more later.

Privacy: anything you put in the Knowledge Base is searchable across every call this Finn handles. Don't put per-caller data here (like account balances or order status) — for that, use a live data lookup webhook instead.

Where to configure

Create New Finn → Knowledge Base tab. Also editable later via View All Finn → [Finn] → Edit → Knowledge Base.

Three sources, one index

You can mix and match three types of source material:

1. Uploaded files — PDFs, DOCX, Markdown, plain text.
2. Pasted text — copy-paste content directly into the editor.
3. Website imports — provide a URL, the platform crawls and indexes pages.

All three feed into a single searchable index per Finn. The Finn doesn't know (or care) where a snippet came from at call time.

Uploaded files

Supported formats

Limits

• Max file size: 25 MB per file.
• Max files per Finn: 100.
• Total indexed size: 500 MB per Finn (varies by plan).

What gets indexed

The platform extracts text, tables, and headings. Images and scanned-only PDFs (no embedded text layer) are skipped — OCR them externally first.

Best practices

• Smaller files, more of them > one giant PDF. The retrieval system performs better on focused documents.
• Strip boilerplate — headers, footers, page numbers add noise.
• Keep one topic per file — easier to update without re-indexing everything.

Pasted text

For quick-and-dirty content that doesn't deserve a file. Limits: 100,000 characters per paste, 10 pastes per Finn.

Use it for:

• FAQ snippets.
• Short product descriptions.
• Daily-changing data (today's promo code, current wait times) — though for truly live data, use a Webhook task instead.

Website imports

Provide a URL. The platform crawls the site and indexes pages.

Configuration

What works well

• Public-facing help centers and FAQ pages.
• Product catalogs with clean HTML.
• Static-rendered marketing pages.

What doesn't

• Sites that require login.
• Heavy single-page apps where content is JS-rendered after page load (use file upload instead).
• Sites that block crawlers via robots.txt (the platform respects it).

How retrieval works at call time

When the caller asks a question:

1. The Finn forms a query from the caller's utterance.
2. The platform searches the index (vector + keyword hybrid).
3. The top-N matching snippets are injected into the LLM prompt.
4. The Finn synthesizes an answer grounded in those snippets.

You can inspect what was retrieved per call in the Data Extractor Sidebar → Knowledge tab. Useful for debugging "why did the Finn say that?"

Tuning retrieval

If the Finn is giving wrong or off-topic answers:

• Add more specific content. If "what's your refund policy?" returns vague answers, add a dedicated refund-policy file.
• Remove contradictions. Two files saying different things confuse retrieval. Audit and reconcile.
• Tighten the welcome message guardrails. Tell the Finn explicitly what's in scope. "You only answer questions about scheduling. Anything else, transfer to a human."
• Re-rank. Premium plans let you set per-source priority — newer files trump older ones.

When NOT to use the knowledge base

Some data shouldn't live here:

• Per-caller data (account balances, order status) — use a Webhook task to look it up live.
• Sensitive data (SSNs, full credit card numbers) — never paste this in. The KB is searchable across all calls.
• Highly dynamic content (prices that change hourly) — also a webhook job.

The KB is for background context that's the same for every caller. Anything caller-specific or time-sensitive should be a live lookup.

Re-indexing

The platform re-indexes:

• Automatically — when you add/remove/edit a source.
• On schedule — for website imports, per the re-crawl frequency.
• Manually — Knowledge Base → Re-index Now. Forces a full rebuild.

Re-indexing typically takes seconds (small KBs) to a few minutes (500 MB).

Cost

Knowledge Base storage and queries are included in your plan up to the size limits above. Exceeding limits requires a plan upgrade — see Billing →.

Next

• Workflows → — when free-form retrieval isn't structured enough.
• Integrations → — live data lookups via webhooks.