Steno

Steno is an open-source, privacy-first AI meeting notetaker built by Skrape Limited and maintained by developer Ruzin Saleem. It runs the entire pipeline — audio capture, transcription, and AI summarization — locally on your device, with zero network requests after install. The project is MIT-licensed, available on GitHub, and targets professionals in government, defense, legal, healthcare, finance, and executive roles who handle confidential conversations.

What It Is

Steno is a desktop application for macOS (Apple Silicon) and Windows (alpha) that replaces cloud-based meeting notetakers with a fully on-device stack. It captures microphone and system audio without joining calls as a bot, transcribes speech in real time using Parakeet TDT v3 or Whisper, and generates structured summaries, key topics, and action items using local LLMs via Ollama. Because every step happens on the machine, recordings and transcripts never touch an external server.

How the On-Device Pipeline Works

The workflow has three stages, all local:

• Record — Captures mic input, system audio, or both. On macOS 14.4+, system audio is captured via a native Core Audio Tap, so no virtual cable or screen recording permission is needed. Both sides of a video call are recorded without any bot joining.
• Transcribe — Parakeet TDT v3 (MLX, Apple Silicon) provides live on-screen transcription with speaker labels ([You] vs [Others]) in 25 European languages. Whisper Large V3 Turbo handles 99 languages including Chinese, Japanese, Arabic, and Hindi in post-stop mode.
• Summarize — Six local models are bundled via Ollama, ranging from Gemma 4 E2B (2B, lightest) to GPT-OSS (20B, most capable). Users can also connect OpenAI, Anthropic, Gemini, AWS Bedrock, or a custom API endpoint using their own key, with data going directly to the provider and no Steno middleman.

Target Audience and Enterprise Positioning

Steno is explicitly designed for regulated and sensitive workflows. The enterprise page covers government, defense, legal, healthcare, finance, and executive use cases. Because recordings never leave the device, the vendor states this architecture supports HIPAA and GDPR obligations by design — no third-party processor handles the audio. The defense page notes it is designed for air-gapped, local-only deployments. The README notes it is sponsored by GitLab Founder's Open Core Ventures, and the vendor states it is trusted by users at AWS, Deliveroo, Tesco, HashiCorp, Rutgers, and the European Union.

Key Capabilities

• Live transcription with speaker diarization during recording
• Auto start/stop detection when a meeting begins or ends
• In-app note-taking merged into the AI summary
• Natural-language Q&A across a single note or the entire library via a Chat tab
• Report templates for custom summary styles per meeting
• Markdown-native output for notes and transcripts, compatible with any knowledge base or AI agent workflow
• A granola-to-steno agent skill for migrating notes from Granola
• macOS Shortcuts integration via stenoai:// deep links for calendar-triggered recording
• 99-language transcription with automatic language detection

Update: Steno 0.6.5

The latest release is v0.6.5, published on 2026-07-26. Notable changes in this release include system audio capture without requiring Screen Recording permission (now requires macOS 14.4+), automatic background updates that install while idle and never interrupt a recording, undo-delete and re-transcribe from the note detail view, and Simplified/Traditional Chinese transcription selection when using the Whisper engine. Windows alpha support (10/11 x64) ships the full pipeline including live Parakeet transcription and system-audio loopback with speaker diarization, though it remains unsigned and CPU-only for summarization. The project has 1,242 GitHub stars and 166 forks as of the latest data.