Skip to content
Veritas AI

Quickstart

From zero to a cited answer in five minutes.

1. Get an API key

Create a key from the dashboard, then export it:

bash
export API_KEY="vt_live_..."

2. Ingest a document

bash
curl -X POST https://api.veritas.example.com/v1/ingest \
  -H "Authorization: Bearer $API_KEY" \
  -F "file=@bank-policy.pdf"

The response includes a job_id. The pipeline parses the PDF (running OCR on scanned pages), chunks it, embeds every chunk, and writes to the vector index. Most documents are searchable within seconds.

3. Ask a question

bash
curl -N -X POST https://api.veritas.example.com/v1/chat \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"question": "What is the minimum balance?"}'

The stream sends a [SOURCES] frame first — the retrieved chunks with file, page, and score — then one frame per token, then [DONE]. Render citations immediately; the answer's [1] markers index into that sources array.

What just happened

  1. Your question was embedded and searched against both the vector index and a BM25 index in parallel
  2. The two rankings were fused with Reciprocal Rank Fusion
  3. A cross-encoder reranked the fused candidates down to the top 5
  4. The LLM generated strictly from those chunks, citing as it went

Next: understand how ingestion works, or jump to the API Reference.

PreviousOverviewNext Ingestion