Three Open‑Source Gems: Local‑First Knowledge Hub, NL‑to‑SQL AI, and Private Speech‑to‑Text

Welcome to the new open‑source project weekly (Nov 3‑9). This edition focuses on knowledge management, data analysis, and privacy, presenting three heavyweight tools.

01 AFFiNE: Knowledge‑Management Operating System

AFFiNE is an open‑source, privacy‑focused, local‑first workspace that combines Notion’s document/database capabilities with Miro’s infinite canvas, allowing seamless switching and collaboration.

Key highlights:

True canvas integration: An infinite canvas that can host rich text, sticky notes, embedded webpages, multi‑view databases, shapes, charts, and any other building blocks.

Local‑first architecture: Data is stored on the local disk by default, offering offline‑first operation; cloud sync is optional, emphasizing self‑hosting, full data ownership, and privacy.

Multimodal AI: Built‑in AI can generate reports, slides, mind‑map summaries, task plans, and even code prototypes.

Tech stack: Open‑source and cross‑platform.

GitHub: https://github.com/toeverything/AFFiNE

02 Vanna: Natural‑Language‑to‑SQL RAG Framework

Vanna is an MIT‑licensed open‑source Python RAG (retrieval‑augmented generation) framework designed to translate natural‑language questions into precise SQL queries and execute them, returning visual results, thereby lowering the barrier for data analysis.

Key highlights:

Core functionality: Converts natural‑language questions to exact SQL, runs them on the database, and returns visualized results.

Extensible architecture: Modular design lets users mix and match LLMs (OpenAI, Gemini, Anthropic, etc.), vector stores (ChromaDB, Pinecone, PgVector, …), and databases.

Privacy and flexibility: Supports local components (e.g., Ollama, ChromaDB) for data privacy or cloud services for convenience; all components are abstracted for easy swapping.

Quick start: Install with vanna, add DDL, documentation, and SQL examples, then query via the ask() method.

GitHub: https://github.com/vanna-ai/vanna

03 Handy: Offline, Privacy‑Focused Speech‑to‑Text Application

Handy is a free, open‑source, fully offline speech‑to‑text app built with the Tauri framework (Rust + React/TypeScript). It runs locally, ensuring that audio data never leaves the device, providing high performance and data security for privacy‑conscious users.

Key highlights:

Fully offline operation: All audio processing and data stay on the device, guaranteeing privacy with no telemetry or usage tracking.

Dual engine support:

Whisper: High‑accuracy, multilingual model (GPU‑accelerated recommended).

Parakeet V3: CPU‑optimized, delivering about 5× real‑time speed on mid‑range hardware.

Convenient recording modes: Supports “hold‑to‑talk”, mode switching, and “always‑on” recording, and can be integrated globally into any text field or application.

Cross‑platform support: macOS, Windows, Linux with automatic GPU detection (Metal, CUDA, DirectML, OpenCL) for acceleration.

GitHub: https://github.com/cjpais/Handy

These three open‑source projects illustrate the latest advances in local‑first design, high performance, and AI integration, offering tangible benefits for developers and knowledge workers.