Knowledge Base (RAG)
Teach your AI agent to answer from your real business information — upload documents or connect Notion, and every reply is grounded in your own content.
What the Knowledge Base Does
The Knowledge Base is how your AI agent answers from your real information instead of guessing. You give it your documents; it reads them, and at answer time it pulls out the most relevant passages and replies based on those — a pattern called RAG (Retrieval-Augmented Generation).
Waslo tells the AI that the Knowledge Base is its authoritative source and instructs it to prefer that content over anything it might otherwise assume. When the answer comes from your documents, the AI cites the source in the chat, so you (and your customer) can see where it came from.
Find it under the AI section in the sidebar → Knowledge Base.
How It Works
Behind the scenes, every document goes through the same pipeline:
- Extract — Waslo pulls the plain text out of your file or Notion page.
- Chunk — the text is split into overlapping passages so related sentences stay together.
- Embed — each chunk is turned into a vector (a numeric representation of its meaning).
- Retrieve — when a customer asks something, Waslo finds the chunks closest in meaning to their question.
- Answer — those chunks are handed to the AI, which writes a reply grounded in them and cites the source.
This all happens automatically on every reply, across every channel. You don't manage any of it — you just keep your documents current.
Supported Sources
| Source | Notes |
|---|---|
| Must contain real, selectable text (see caveats below) | |
| Word (DOCX) | Standard documents |
| TXT | Plain text |
| Markdown (MD) | Great for structured FAQs |
| CSV | Rows of structured data (e.g. FAQ pairs, spec tables) |
| Notion | Connect a page via an internal integration token |
Each uploaded file has a 10 MB size limit.
Adding Documents
- Open Knowledge Base in the sidebar.
- Click to upload a file (PDF, DOCX, TXT, MD, or CSV).
- Wait for the status to move from processing to ready — this can take a moment while Waslo extracts, chunks, and embeds the text.
- Once ready, the document is live: the AI starts using it on the next reply.
You can remove a document at any time; its content stops being used immediately.
Connecting Notion
Notion lets you keep your source of truth in one place and have Waslo stay in sync with it.
- In Notion, create an internal integration and copy its token.
- Share the specific page (or database) with that integration — this is the step people miss. If the page isn't shared with the integration, Waslo can't read it.
- In Waslo, connect Notion with the token and point it at the page.
- Waslo walks the page's blocks, extracts the text, and indexes it like any other document.
A daily re-sync keeps the Notion content fresh, so edits you make in Notion flow into the Knowledge Base without re-uploading.
Notion gotcha: a page whose content is only an embedded file (for example an embedded PDF or Google Doc) has no readable text of its own — the embed points elsewhere. Put the actual text on the Notion page, or upload the file directly instead.
Document Caps
| Plan | Max documents |
|---|---|
| PAYG | 50 |
| Growth | 200 |
If you hit the cap, remove documents you no longer need or upgrade to Growth for more headroom.
Scoping Documents to Specific Numbers
If you run more than one WhatsApp number (for example one per branch or per brand), you can scope a document so it only applies to certain numbers. A document with no scope applies to your whole organization by default. Scoping is useful when a price list, policy, or menu differs between locations and you don't want one branch's document leaking into another's replies.
It's Text-Based — Read This
The Knowledge Base works on text. If a document has no extractable text, it can't be indexed. The two common cases:
- Scanned-image PDFs — a PDF that is really just photos of pages (a scan or a screenshot export) has no selectable text, so nothing can be extracted.
- Notion pages that are only an embed — as described above, the readable content lives in the embedded file, not on the page.
When this happens, the document is marked failed with an actionable message, and you get a notification. The fix is simple: re-upload a text version — export the document with real text, run the scan through OCR first, or paste the content in as Markdown/TXT.
Prices Belong in the Catalog, Not the Knowledge Base
Important rule for owners: product prices belong in the Product Catalog (Growth), which is the authoritative source for prices and availability. Do not duplicate prices in both the Catalog and the Knowledge Base.
If a price lives in two places and one is out of date, the AI can quote the wrong number. Keep prices in the Catalog and keep everything else — policies, hours, FAQs, how-tos, service descriptions — in the Knowledge Base.
Best Practices
- Write for questions. Structure content as clear FAQ-style entries ("How long is the warranty?" → answer). Retrieval works best when a chunk reads like a direct answer.
- Keep it current. Outdated hours or policies produce confidently wrong replies. Review documents when your business changes.
- One fact, one place. Avoid contradicting yourself across documents — conflicting passages confuse retrieval.
- Be concise and specific. Short, focused sections retrieve better than long, rambling ones.
- Use the Catalog for products. Prices, stock, and photos go in the Catalog, not here.
Troubleshooting
| Symptom | Likely cause | Fix |
|---|---|---|
| Document stuck / marked failed | No extractable text (scanned PDF, embed-only Notion page) | Re-upload a real text version or run OCR first |
| AI ignores a document | Still processing, or removed | Confirm status is ready |
| AI quotes the wrong price | Price duplicated between KB and Catalog | Remove the price from the KB; keep it in the Catalog |
| Notion page won't index | Page not shared with the integration | Share the page with your Notion integration |
| Answer isn't found | Content buried in long prose | Rewrite as a focused, FAQ-style section |
FAQ
Does the AI read the whole document on every reply? No. It retrieves only the most relevant passages for the customer's specific question, then answers from those. That keeps replies fast and focused.
What file formats can I upload? PDF, Word (DOCX), TXT, Markdown (MD), and CSV — up to 10 MB each.
Why did my PDF fail to index? Almost always because it's a scanned image with no selectable text. Export or re-create it as a text PDF (or OCR it), then re-upload.
How fresh is my Notion content? A daily re-sync pulls in changes. For an urgent edit you can reconnect the page to re-index sooner.
Should I put my price list in the Knowledge Base? No — prices go in the Product Catalog (Growth), which is authoritative for pricing and availability. Keep everything else in the Knowledge Base.
Does the customer see the source? When a reply is grounded in your documents, the AI cites the source in the conversation.
Next Steps
- AI Agent Configuration — personality, classification, handoff
- Voice & Media Understanding — voice notes, photos, and video in and out
- Getting Started — connect a channel and go live