Gremble-io/Detto

What it solves

Detto is a local-first macOS application designed to capture and transcribe voice data without sending audio or text to the cloud. It solves the problem of data lock-in and privacy concerns associated with cloud-based transcription services by providing a system-wide tool for meeting capture, voice memos, and dictation that runs entirely on Apple Silicon hardware.

How it works

Detto utilizes a dual-stream audio capture system to separate microphone and system audio (for calls), and processes this data using on-device AI models. It uses the Parakeet-TDT v3 model for Automatic Speech Recognition (ASR) on the Apple Neural Engine and Llama 3.2 3B (via MLX) on the GPU for text refinement to fix punctuation and capitalization. The output is saved as plain Markdown files with YAML frontmatter, making them compatible with knowledge bases like Obsidian.

Who it’s for

It is primarily aimed at consultants, researchers, and professionals who maintain a "second brain" or personal knowledge management system (like Obsidian) and require high-privacy, offline transcription and dictation tools that integrate directly into their file-based workflows.

Highlights

• Local-first architecture: No accounts, no servers, and no audio saved to disk.
• Channel-based speaker separation: Distinguishes between the user ("You") and others ("Them") during calls.
• Vault-native output: Transcripts are saved as .md files with structured YAML metadata.
• Client Briefing: Scans local markdown files to provide pre-call context and embed client data into transcripts.
• System-wide dictation: A hotkey-driven tool that injects transcribed text directly into any active application.
• Conferencing app detection: Automatically tags transcripts based on the active app (e.g., Zoom, Teams, Slack).
• Multi-language support: ASR auto-detects 25 European languages.