Upload a plain text file (up to 1 MB), then ask questions about it. The tool splits your text into chunks, finds the most relevant ones by keyword matching, and sends them alongside your question to an AI model. You see the answer and exactly which parts of the file were used — no server-side storage, no vector database, no API key needed.
Do not upload files containing personal, confidential, or sensitive information (passwords, financial records, medical data, private documents). When you ask a question, the relevant parts of your file are sent to a third-party AI service (chat.velstech.net) to generate an answer. See the Privacy Policy.
Drop a .txt file here
or click to browse — up to 1 MB
All processing happens in your browser. The file content is sent to chat.velstech.net only when you ask a question (the relevant chunks + your question). No data is stored on any server. Chat history is kept in session storage and cleared when you close the tab.
How the RAG works
This is a simple keyword-matching retrieval-augmented generation (RAG) pipeline — no embeddings, no vector store.
1. Chunking
The file is split into overlapping chunks of ~500 words each (~50 word overlap). Each chunk retains its approximate position in the document.
2. Scoring
Your question is tokenized (lowercased, non-alphanumeric stripped). Each chunk is scored by counting how many of the question's tokens appear in it, with a frequency bonus for repeated terms.
3. Retrieval
The top 3 scoring chunks are selected as context. If no chunk has a score above 0, the question is sent without context.
4. Generation
The context chunks + your question are sent to the AI model at chat.velstech.net (Llama 3.1 8B via Cloudflare Workers AI). The model is instructed to answer based only on the provided context.