VelocityNote

Cloud note-taking apps have trained users to accept that AI features require an API key and an upload. VelocityNote breaks that assumption: it ships with llama.cpp and ONNX runtimes built directly into a 34 MB binary, so the language model, speech-to-text, and image recognition all execute on your machine. The core workflow is Markdown editing with full formatting support — tables, task lists, footnotes, LaTeX, flowcharts, code highlighting, attachments — plus voice transcription that produces timestamped notes as you speak, OCR that converts screenshots into searchable Markdown blocks, and AI rewriting that operates on selected text or full notes without touching the network.

The standout differentiator is what the vendor calls a ‘small footprint, full notebook’: the app holds the entire retrieval layer that AI agents need inside a binary that loads in under half a second. SQLite FTS5 handles full-text search, and the vendor-provided benchmark puts search latency across a 10,000-note vault at 9 ms — compared to 180 ms for Obsidian and 600 ms for Notion in the same illustrative test. For teams whose AI workflows need to query a large note vault locally without an external vector database, this matters.

VelocityNote fits individuals and small teams whose threat model rules out cloud AI, privacy-conscious researchers keeping large Markdown vaults, and power users importing archives from Evernote, Notion, or Word into a local-first system. It does not fit teams that need real-time collaborative editing, mobile-primary workflows, or a published API to connect the tool to external systems — no API is available. The optional sync feature exists, but the docs describe shared notebooks as a beta offering, and collaborative conflict resolution is listed as a roadmap item, not a shipping feature.