Add your sources
Drop in PDFs, websites and your own notes. Istor extracts the text and keeps the originals in a library you own.
Istor is a local research notebook. It runs entirely on your machine: add papers, websites and notes, ask in your own words, and read answers where every claim points back to the passage it came from.
Local model, local database, no telemetry. The only network traffic is the fetches you trigger.
Why Istor
Most tools hand you a paragraph and hope you trust it. Istor answers from your own sources, and every claim carries a citation you can open.
The citations in this preview work. Try them.
Example
What did the 2019 field survey find about habitat fragmentation?
Species richness declined most sharply in , while core forest interiors stayed across the seven-year window.
Field survey report, 2019 · page 14
Patches below ten hectares lost on average 31 percent of recorded species within four seasons, the steepest decline recorded in the study.
Field survey report, 2019 · page 17
Interior plots beyond 500 metres from any edge showed no significant change in recorded richness between 2012 and 2019.
Full-text search finds the exact phrase. Semantic search finds the passage that means what you asked, even when the wording is completely different.
transect methods for dawn bird counts
Field notes, 2021 · page 3
Counts followed fixed transects, beginning thirty minutes before sunrise.
Matched the words
Field survey report, 2019 · page 9
Transects were walked within one hour of sunrise, at a fixed pace, on three consecutive mornings each month.
Matched the meaning
2 passages · 3 sources · searched on this machine
Ask something your library cannot support and Istor does not improvise. It names what it could not find and what it would need, so you always know where the edge of your evidence is.
Why did the survey stop at the eastern river?
Your sources do not say. Three papers mention the river, but none explains the decision.
No answer offered
→ Fetch the survey methodology from the web?
The loop
Drop in PDFs, websites and your own notes. Istor extracts the text and keeps the originals in a library you own.
Ask in your own words. Istor reads the library and answers from what your sources actually say.
Every claim carries a citation. Click one and the exact passage opens beside the answer.
Inside the notebook
Your library, your model and every citation, in one window. Nothing is a tab you have to hunt for.
The notebook
PDFs, web pages and plain notes land in one library. Duplicates are detected and merged, so the library stays clean as it grows.
Full-text search finds the exact words. Semantic search finds the idea behind them, so a question in plain language still reaches the right passage.
Pick a model from the list in settings. Swapping takes effect immediately, with no restart and nothing leaving the machine.
Watch the reasoning form live while the answer streams. When a claim matters, you can see the path it took to get there.
Write alongside your research. Notes live in the same library and are searchable together with everything you imported.
Export the whole library as a zip any time. One file carries your sources and notes, ready to move or restore.
Try it here
Three questions from one small sample library. Pick one and watch the loop run. It all happens in your browser; the app runs the same loop on yours.
What did the 2019 field survey find about habitat fragmentation?
read the library · 3 sources · checked every claim
Species richness fell most sharply in small habitat patches, while core forest interiors stayed relatively stable across the seven-year window.
Field survey report, 2019 · page 14
Patches below ten hectares lost on average 31 percent of recorded species within four seasons, the steepest decline recorded in the study.
A miniature of the real loop. The app runs it on your library, with your model, and cites your pages.
Who it is for
Drop a semester of papers into one library and ask across all of them. Every borrowed sentence keeps the page it came from.
Survey notes, reports and transcripts stay in their original files. Semantic search finds the passage however you phrase the question.
Draft beside your sources. Borrowed facts stay pinned to their pages, so checking a claim never means losing your place.
Privacy
Istor runs entirely on your machine.
Inference runs on your own hardware through open models. Nothing is sent to a server.
Your sources live in a SQLite database on your own disk. Back it up, copy it, carry it.
No accounts, no analytics, no usage reports. There is nothing to opt out of.
The only requests Istor makes are the fetches you ask for.
No. Research works fully offline. Istor only touches the network when you ask it to fetch a web source.
No. Retrieval, prompting and inference all happen on your machine. Nothing leaves it unless you trigger a fetch yourself, and the whole library exports as a zip any time. If you are weighing tools on privacy, here is what to check in any of them.
A GTX 1650 class GPU runs the models comfortably. There is no cloud requirement, so the hardware you have is the hardware it uses. The details are on can you run a research AI on a GTX 1650.
Yes. Pick your model in settings, and a swap takes effect without a restart.
Istor says so, and names what a real answer would need. It will not produce one your sources do not support.