for developers who'd rather ship
A working template that lets anyone chat with their own PDFs — chunking, embeddings, retrieval, and grounded answers with citations, already wired up. Download it, drop in your keys, run it.
You can. The pieces are all open source. This saves you the week of getting the unglamorous parts right.
Splitting documents without cutting sentences mid-thought, tuned overlap so context doesn't get lost at the edges.
Fast similarity search over your documents via FAISS, wired to real embeddings — not a keyword search pretending to be smart.
Prompting that keeps answers tied to your documents, with the model told to say when it doesn't know, and to cite its source.
One folder. No account, no subscription, no hosting required.
This is code you run yourself, not a hosted app. Be sure this fits before purchasing.
Runs on Claude specifically — not ChatGPT, not a Claude.ai subscription. You'll need a funded Anthropic developer account. Pay-as-you-go, typically fractions of a cent per question.
Used for embeddings. Free tier covers this comfortably for personal or small-team use — no cost expected here.
You'll run a few Python commands to set it up. Setup docs walk through each step, but this isn't a no-code tool.
One payment. The code is yours — run it, modify it, use it on client work.