Skip to content

note-lm v0.1.0 - Research notebook with verifiable evidence

Choose a tag to compare

View all tags
@Gjusev Gjusev released this 29 Sep 16:10
· 14 commits to main since this release
v0.1.0
c118880

First packaged release of note-lm: a local-first research and study notebook for Windows.

Download

  • note-lm_0.1.0_x64-setup.exe (~99 MB) - NSIS installer, Windows 10/11 x64. No Node.js, Python or any runtime needed: everything the app uses is packaged or app-managed. Verify with the .sha256 file.

What you can do

  • Import PDFs, web URLs, CSV tables, audio and video - everything becomes an immutable, versioned source.
  • Ask questions with local AI (llama.cpp - chat and embedding models download from Settings; Whisper speech-to-text installs itself on first use) or bring your own API provider (OpenAI, OpenRouter, or any OpenAI-compatible endpoint; keys live in the OS keyring, never in files).
  • Save claims anchored to the exact source version, page or time range - open any citation and see the original.
  • Re-import a document as a new version: the app deterministically flags which claims need review (moved or vanished quotes), you accept or reject, history is kept.
  • Run reproducible calculations over CSV versions - strict validation, never LLM arithmetic.
  • Evidence matrix: claims x sources grid showing linked evidence, pending review, and honest "not found in the recorded search".
  • Export research as portable packages (hash-verified, with or without originals) and restore them anywhere.
  • Everything runs inside the one installed program: workers persist with pause/resume/cancel and recover from crashes (proven by automated tests that kill the engine mid-work).

Verification

This release was built by a 19-gate automated pipeline: unit suite (343), engine e2e, clean-PATH installed smoke, kill-and-recover gates, and full walkthroughs (research, revision, calculation, voice) driven from the installed package. Quality measurements (retrieval benchmarks, speech WER, change-review precision) live in the repo under eval/.

Known limits: no code signing yet (SmartScreen may warn), in-app PDF page rendering shows the stored original, and the first-run model downloads need internet.

Assets 4
Loading