Document Chat API with Citations

How it fits

Built on extraction, not beside it.

Already using FileTag or extraction? Chat works on that same data.

1. Extract

Gemina reads each document with FileTag, ready-made invoice fields, or a template you define.

2. Index

Turn on indexing, and every extraction from then on can be searched and asked about. Earlier documents need a one-time backfill.

3. Ask

Ask in Gemina's web app, over the API, or in a chat inside your own product. Each answer cites its sources.

How extraction works

Grounded answers

Every answer carries its sources.

Gemina routes the question to the right engine — a filter, a search, or an exact aggregation — and answers from what it found. Amounts come from the database, not from the model.

Each answer returns the document IDs it used, so your UI can link straight to the source.

How the totals work

A Document Assistant conversation: a question about telecom expenses, then two follow-ups about the supplier's support phone and email, with a source-document citation chip under every answer

# Grounded natural-language Q&A over your document collection
curl -X POST https://api.gemina.co/api/v1/chat/query \
  -H "X-API-Key: your-api-key" \
  -H "Content-Type: application/json" \
  -d '{ "message": "How much did we spend at Acme Catering this quarter?" }'

{
  "answer": "You spent a total of 18,450 ILS at Acme Catering this quarter, across 12 invoices.",
  "citations": ["5f2b7c1e-2b6b-4f0e-9a1c-4d3f2e1a0b9c"],
  "intent": "aggregation",
  "confident": true
}

What people ask

Business questions, not search queries.

These are the questions your users already ask each other. Now they can ask the documents.

How much did we spend with Supplier X?
When did we last buy this product?
How many units did we purchase?
Which invoices are still unpaid?
What did we pay for cleaning supplies in July?
Show me everything from this vendor this quarter.

Embed it

A drop-in chat UI, or your own.

Ship the component as-is, or call the API and build your own interface.

An embedded Document Assistant chat returning an exact $900.24 total with source-document citation chips and a categorized expense breakdown

The React component

@gemina/elements gives you a chat component with citations, confidence handling and right-to-left support built in.

Give it your own title, intro copy, and header color.

Scoped session tokens

Your server exchanges its API key for a short-lived token scoped to one end user. The browser never sees your key.

The token can read and chat, but can never purge history.

Chat history

Past conversations are kept as sessions you can list, reread and permanently purge, each with an auto-generated title.

Your retention window sets when a session is deleted.

Put it inside your product See the component

Worth knowing

What to plan for.

Indexing is opt-in, and not retroactive

Turn indexing on and every extraction from then on becomes searchable and chattable. Documents processed earlier need a one-time backfill.

Chat runs over REST and SDK

Agents get search, aggregation and indexing over MCP. Chat runs over the API — it carries conversation state that a stateless tool call doesn't.

Chats are included per plan

Each plan includes a monthly chat allowance with per-chat pricing beyond it. Search itself is never metered.

See what your plan includes

Let your users ask.

Start with trial credits. No payment card is needed.

Start Free Trial Read the Docs

Upload your documents. Then ask them anything.