stenolabs/stenoai

What it solves

Steno is a privacy-first AI notepad designed for confidential conversations. It eliminates the need for cloud-based transcription and summarization services, ensuring that sensitive data—such as government, defense, or C-suite meetings—never leaves the user's local machine.

How it works

Steno records system and microphone audio, then processes it locally using a combination of AI models. It uses Parakeet TDT v3 (via MLX on Apple Silicon) for real-time transcription with speaker labels and Whisper for post-recording transcription of a wider range of languages. For summarization and Q&A, it integrates with Ollama to run local LLMs (like Gemma 4 or Qwen 3.5) on-device. Users can also optionally connect their own cloud-based LLM endpoints (OpenAI, Anthropic, AWS Bedrock) if they prefer.

Who it’s for

Professionals handling highly confidential data, including those in government, defense, and executive leadership, as well as anyone seeking a private, local alternative to AI meeting assistants.

Highlights

• 100% On-Device Processing: Recordings, transcripts, and summaries stay local to prevent data leaks.
• Live Transcription: Real-time text generation with [You] vs [Others] speaker attribution.
• System Audio Capture: Native capture of both sides of virtual meetings without requiring virtual cables or screen recording permissions (macOS 14.4+).
• Local Q&A: A chat interface to query specific notes or the entire meeting library using local LLMs.
• Knowledge Base Integration: Syncs notes to Obsidian as Markdown and supports custom report templates.
• Agent Skills: Includes a /steno skill for AI agents to read and act upon local meeting notes.