What it is
A workspace and a retrieval service in one. You upload documents, Contexta chunks and indexes them, and then three things read that index: the chat, the automatic brief on every document, and an API key you can call from your own code.
About
A small product for reading your own documents
Contexta indexes PDFs and DOCX files so you can ask them questions and get answers that point back at the page they came from.
How an answer is built
Your question is embedded and also searched as keywords; the two result lists are fused with reciprocal rank fusion so neither arm can win alone; the surviving chunks are budgeted into a prompt; and the model answers from that context only. Each citation carries the document, the page and the section heading it came from.
What it refuses to do
It does not train on your documents, it does not fill a gap with general knowledge and present it as a source, and it does not show a relevance number it cannot back with a chunk. When retrieval is thin, the answer says the document does not cover it.
How it is run
An independent product, hosted on infrastructure we operate ourselves: a Next.js front end, a FastAPI service, a Python indexing worker, Postgres with row-level security, and a vector store we run. Prices are whole dollars and the limits are published on the pricing page.
The stack, in plain terms
- Next.js 15 front end, deployed on Vercel
- FastAPI service plus a separate Python indexing worker
- Supabase Postgres with row-level security on every table
- Qdrant as the vector store, on our own server
- An embedding model through OpenRouter, DeepSeek for answers
- Stripe for subscriptions