Rajan Verma

Private retrieval · Demo

Ask your files. They never leave your disk.

I put retrieval next to your PDFs so your product can ask an LLM you own. Chunks and answers stay on that disk. Your engineer gets a private query API, a token only your app holds, and a written handoff. The screen stays yours.

See live example

How the install works

  1. 01 Your documents stay on your machine

    Manuals, policies, and tickets are not uploaded to a hosted chat product. Vectors sit on the same disk as the files.

  2. 02 I chunk and embed where the files already are

    Ingest on that machine. Qdrant (or your store) holds the chunks. Nobody trains on your corpus.

  3. 03 Retrieve, then generate — one private path

    Embeddings plus the same on-prem LLM you already run. Your app POSTs a question and gets an answer plus the passages used.

  4. 04 Nobody logs into the model

    Only your app holds the Bearer token. Staff do not get a chat login.

  5. 05 I hand the machine to you in writing

    Runbook, health probe, how to re-index, how to stop. Your engineer is not stuck with my laptop.

  6. 06 Your engineer integrates — one POST

    A few lines in your product. The example below is a fictional prop so you can see the click. It is not the SKU.

What you get

A token site, paste-ready snippets, and a written handoff

After the install your engineer mints the secret, copies a query call into your product, and keeps a one-page runbook — URL, health, re-index, how to stop. The demo is a working mock: fake names, fake token, same click.

Open the token demo

Live example

Example · live · ask files

Folio

A product team keeps SOPs as PDFs on a share. The same question keeps arriving: lockout steps, a part number, which form to file. Today someone opens the PDF, copies a page, and pastes it into a public chat. The file has left the building. Fictional product. Not a real firm. No “hours saved” number.

With private retrieval their product POSTs the question to a query API on their server. They get an answer plus the passages used. The PDFs never leave the disk. Open the live query: the corpus here is public-domain NASA and RFC text as a stand-in. A paid job uses your files.

Open live Folio query

Contact

Let’s keep the files on your disk

If manuals or tickets cannot leave your server, write. Tell me where the files live, whether you already run an LLM, and the date. I come back with a query API, a token, and a written handoff. You keep the product UI.

Live query is a thin add-on. The product is POST /query with a Bearer token. Public-domain stand-in corpus only.