Ask your documents.
Upload a policy, a manual or a price list, then ask questions in plain language. The assistant answers only from your files and shows where each answer came from.
Public demo: your files stay in a private workspace that only your browser can see, and they are deleted automatically after 30 minutes. The original file isn't kept: it is held in memory (or a private temporary file) only while its text is extracted, then discarded. Your browser also keeps the chat in this tab until the timer ends, when it is cleared. Please don't upload confidential or personal data.
2 · Ask a question
Suggested questions (switch on the sample documents first)
Ask a question about your files — the answer appears here, with the passages it came from.
Answers come from an AI model and can be wrong. Check the cited sources before you rely on them.
How it works
- Extract and chunk. Your file's text is split into overlapping passages that keep their page number and section heading.
- Embed and index. Each passage becomes a vector stored in PostgreSQL with pgvector, next to a full-text index.
- Hybrid search. Your question is matched by meaning (vectors) and by keywords (full text), and the two rankings are merged.
- Grounded answer. The model only sees the top passages, must cite them as [1], [2]… and says so when the answer isn't in your files.
In a real deployment the same system runs on your own server, and it can use a private model on an NVIDIA RTX workstation or a DGX Spark so your documents never leave your office.