Ask questions and get answers you can trace back to your own notes.
Upload your documents into a project, then chat with an assistant that answers only from what you gave it — every claim links back to the passage it came from.
Physics 201 · Chats
Sources
Invalidate on write, not on read — event-driven expiry avoids stale reads.
Short TTLs beat manual purge logic for anything derived from a source table.
LRU eviction handles memory pressure, not correctness.
A simple workflow that stays out of your way.
- 01
Create a project
Start a dedicated space for a topic, client, or course — its documents and chats stay isolated from every other project.
- 02
Upload your documents
Add .md and .txt files. Each one moves through pending, processing, and ready as it's parsed, chunked, and embedded.
- 03
Ask questions
Chat with an assistant that answers only from what you uploaded, and cites the exact passage behind every claim.
Built around one idea: an answer is only useful if you can trace it.
Grounded in your documents
Answers come only from what you uploaded — not from general model knowledge you can't verify.
Citations you can click back to
Every claim links to the exact passage it came from, so you can check it in seconds.
Projects keep knowledge isolated
A client's notes never bleed into a course's notes. Each project is its own knowledge base.
Processing status, live
Watch each document move from pending to processing to ready, so you know exactly what the assistant can see.
Create your first project.
Upload a few documents and ask your first question in under a minute.