What you start with
- A folder of papers, reports and saved articles
- Your own notes, in the pages where you were thinking
- An export or two that nobody has looked at properly
- A brief due to someone who will ask “says who?”
Use cases · The research pile
The hard part of research is not reading. It is holding twenty things in your head at once, and being able to say where a claim came from three weeks later when somebody challenges it.
Read everything in this folder and my notes on it, and write me a brief on what they agree and disagree about — with the source for each point.
One message. LumenAgent plans the run, asks before anything leaves your machine, and shows each step as it happens.
PDFs, documents, spreadsheets and your own note pages, from your disk. Long documents are read in passes rather than skimmed.
What the sources agree on, where they conflict, and what none of them addresses — which is usually the most useful of the three.
Each claim carries where it came from, so the brief survives being questioned.
Where sources disagree on a number, they go side by side in a sheet rather than being averaged into a sentence.
Nothing rewrites the pages you were thinking in. The brief is a new document; your notes are untouched.
Everything is in the same full-text search afterwards — your notes and the sources together, with wiki-links and backlinks between the pages you connected by hand.
Real files, in the formats other people expect. Open any of them and keep working — nothing here is a screenshot of an answer.
Research is rarely a single event. If you track a subject continuously, save the run and point it at the folder again when it grows — the brief updates instead of being rewritten. Notes made along the way stay linked, so the reasoning is still there next quarter.
Full-text search across your notes and library is instant and local. There is also semantic search — asking in your own words rather than by keyword — which needs an account and a network connection, because the embeddings are not computed on the device yet.
Notes are encrypted at rest using your operating system’s keychain, and the notes index is held in memory so the ciphertext never lands on disk in searchable form.
No. AI output goes into new documents. Your note pages are edited by you.
Yes — wiki-links between pages, with backlinks generated automatically, so you can walk the connection from either end.