OPEN SOURCE · MIT · ZERO-DEPENDENCY CORE

AI Notes

Source-grounded, bilingual, local-first AI note taker. Turn documents, web pages, recordings, videos and YouTube links into structured notes where every bullet cites the exact source chunk it came from — so you can verify, edit and reuse notes instead of trusting a black box.

View on GitHub → Find us on Product Hunt Try Lynote AI Notes

Why Source-Grounded?

AI note tools usually give you a summary you cannot check. Lynote Notes is built around traceability.

• Citations as a ContractEvery bullet carries a source citation. Providers return chunk ids; the core resolves them and drops anything that does not exist — a hallucinated citation cannot leak into a note.
• Offline by DefaultThe default extractive provider needs no model and no network. Configure any OpenAI-compatible endpoint (Ollama, vLLM, OpenAI) for generative notes, with automatic fallback.
• Bilingual by DesignCJK unigrams + bigrams, sentence splitting on 。！？； — English and Chinese work out of the box.
• Local-First WorkspaceOne workspace is one SQLite file. No server, no account, inspectable with standard tools.
• Media TimestampsAudio and video citations point to HH:MM:SS, not just "somewhere in the file".
• Reuse Built InExport Markdown for reading or Anki TSV for memorising — straight from the same cited note.

Quickstart

Python 3.9+, works offline.

git clone https://github.com/lynote-ai/lynote-notes.git
cd lynote-notes
pip install -e .

lynote-notes add lecture.md
lynote-notes add https://example.com/post
lynote-notes note --title "Weekly reading"
lynote-notes ask "What did we decide about pricing?"
lynote-notes export --note note_xxx --format anki --out cards.tsv

How It Works

Five small stages, one traceable pipeline.

1. Step 1IngestText/Markdown, web pages, PDF, DOCX, YouTube captions, audio and video — optional backends stay behind extras.
2. Step 2ChunkParagraph-aware splitting with overlap; media keeps timestamps.
3. Step 3NoteA provider drafts sections from chunks; the core resolves every bullet to a real citation.
4. Step 4AskBM25-lite retrieval finds the evidence; answers cite the chunks they came from.
5. Step 5ExportMarkdown for reading, Anki TSV for memorising.

Limitations

• Extractive DefaultThe default provider selects and organises sentences from your sources rather than rewriting them. Use an LLM provider for more fluent notes.
• Lexical RetrievalBM25 matches words, not meanings. Paraphrased questions can miss relevant passages; embeddings are on the roadmap.
• Transcription QualityDepends on the ASR model size and audio quality — check important details against the source.
• No Live CaptureNo meeting bots by design: export the recording and upload it.

FAQ

How is this related to Lynote AI Notes?

This Space presents the open-source core (MIT, on GitHub). Lynote AI Notes is the hosted product at lynote.ai/ai-notes, which adds a workspace UI, more formats and managed AI. The CLI and library here work standalone.

Do I need an API key?

No. The default extractive provider runs offline. An OpenAI-compatible endpoint is optional for generative notes.

Does it work with Chinese content?

Yes — tokenization and sentence splitting are bilingual by design, and notes can be generated from mixed-language workspaces.

How do I verify a citation?

Every citation names the source, a locator ("chunk 4" or "00:12:30") and a snippet. Open the source at that locator and compare — that is the point of the design.

This Space is a static demo page for the open-source project. The interactive Gradio demo (paste text → cited note → ask questions) is kept ready in this repository and will be enabled once the account has PRO hosting. Full product: lynote.ai/ai-notes. AI notes can miss context or nuance — always check important details against the cited source.