Ask any company document, in a sentence

Every team has the same problem. The answer is in a doc. Nobody can find the doc. So the question gets asked again on Slack, and someone (usually the new hire) gets nominated to track it down.

Kin solves this with a one-time setup: point it at a folder in your Google Drive, and from that moment on, you can ask questions in plain English and get answers grounded in your actual files.

How indexing works

1. Open Documents in your Kin dashboard.
2. Paste a Drive folder URL.
3. Kin walks the folder, reads every supported file (Google Docs, Sheets, Slides, PDFs, DOCX, text/CSV), splits each into ~2,000-character chunks, and stores 768-dimensional embeddings in your private vector store.
4. Each file flips from indexing… to indexed as it finishes. You see a counter of chunks total.

That's it. There's nothing else to configure.

Asking questions

The next time you chat with Kin — on the web or in Telegram — questions that touch your indexed files automatically trigger search_documents. The agent retrieves the most semantically similar chunks, reads them, and answers with a citation:

You: What does our remote-work policy say about Fridays?

Kin: According to HR-Handbook.pdf (chunk 4): Fridays are designated focus days. Employees may opt into a 4-day onsite week with Friday remote, subject to manager approval. Core hours remain 10am–3pm in your local timezone.

If the answer isn't in your indexed documents, Kin tells you that directly instead of making something up.

What "grounded" really means

Two technical details matter:

• Embeddings are semantic, not keyword. Asking "are we hybrid?" matches the policy paragraph above even though it never uses the word "hybrid."
• Each chunk carries a citation back to its source file. You can click through to the original document in Drive at any time.

Costs and limits

Indexing is one-time per file (and free at this scale — Google's text-embedding-004 model is on the free tier for normal usage). The vector store lives in your Supabase project. We cap each document at 200K characters to keep things bounded.

Re-indexing a file replaces its old chunks — no duplicates accumulate.

Best fits

• Onboarding docs ("what's our PTO policy?")
• Sales playbooks ("how do we handle objections about pricing?")
• Technical runbooks ("what's the rollback procedure for the payment service?")
• Meeting archives ("did we decide on the Q3 OKRs yet?")
• Anything that should be answered the same way every time, by anyone who asks.

Start with one folder. Watch how often people stop asking you the same question.