AI 服務暫時不可用，以下為來源正文，待恢復後補全翻譯。
Hi HN — I built VelocityNote, a compact, Markdown cross-platform notebook. The desktop app is under 100 MB. It starts quickly and remains responsive under memory pressure. VelocityNote has a built-in, pluggable AI system that can run local models: - It is a local LLM app: you can use a llama.cpp-based local agent to summarize notes and tasks. More features, including translation, are coming. - It is an intelligent screenshot app: built-in OCR extracts text, while vision models can describe your images. - It is a dictation app: with Silero VAD and speech-to-text, you can record and transcribe hours-long meetings entirely on your device. - It is an app for managing documents: it supports importing a variety of file formats, including PDF and Word, and migrating data from other notebook databases. With local models, your data always stays on your device, with no external provider or per-token limits. Under the hood, notes are stored in SQLite. Full-text search makes searching large notebooks fast. Everything remains available offline and can be exported as standard Markdown. VelocityNote also exposes an MCP server and a command-line interface, allowing it to serve as a local second brain for both you and your AI agents. Agents can use ultra-fast full-text search to find notes, then read, create, and update them through explicit tools instead of scraping the UI or directly modifying the database. VelocityNote is cross-platform and currently available for Windows and macOS. Download: https://velocitynote.app/download Releases, issues, and feedback: https://github.com/VelocityNote/velocitynote