Release · General Availability
Oyma v1.0.0: General Availability & On-Device Engine
General Availability release featuring ScreenCaptureKit loopback, on-device Whisper Turbo, SQLite FTS5 search, and live Obsidian vault interop.
Release Highlights
- ScreenCaptureKit CoreAudio loopback with zero recording bots
- On-device Whisper Large v3 Turbo transcription accelerated by Apple Neural Engine
- Local SQLite FTS5 full-text search indexing with sub-10ms query latency
- Live bidirectional Obsidian vault mounting with .obsidian configuration isolation
- Local Ollama inference support for private LLM meeting summaries
- Hardware-level local-only network isolation mode with auditable egress ledger
Release Overview: Production-Ready Local-First Intelligence
Oyma v1.0.0 marks our General Availability milestone. This release delivers a fully integrated, local-first workspace for macOS that pairs a high-performance CommonMark and GitHub Flavored Markdown editor with on-device meeting recording, speech transcription, and local AI synthesis.
Every subsystem has been engineered from the metal up to guarantee privacy: zero audio packets leave your Mac, no third-party recording bots join your video conferences, and your intellectual property resides exclusively in plain Markdown files on your local filesystem.
Core Capabilities Shipped in v1.0.0
1. Botless System Audio Capture via ScreenCaptureKit
By integrating directly with macOS 14.2+ ScreenCaptureKit APIs, Oyma captures pristine loopback audio from Zoom, Google Meet, Microsoft Teams, Slack Huddles, and Cisco Webex without requiring kernel extensions, virtual soundcard drivers, or intrusive cloud meeting bots. Audio capture operates host-first, preserving the professional decorum of your executive briefings and candidate interviews. Explore the ScreenCaptureKit audio guide and our Zoom setup documentation.
2. On-Device Speech Transcription with Whisper Turbo
Our speech-to-text pipeline runs OpenAI’s Whisper Large v3 Turbo model locally using Apple Silicon CoreML optimizations. The model achieves an outstanding 0.14x Real-Time Factor (transcribing an hour of dialogue in under 9 minutes) with over 98% accuracy across 13 languages. Learn more about our Whisper transcription benchmarks and model quantization in the Whisper model documentation.
3. Sub-10ms Full-Text Search with SQLite FTS5
To support research archives containing tens of thousands of notes, v1.0.0 embeds an in-process SQLite FTS5 search engine. The index uses Porter token stemming and BM25 relevance scoring to deliver instantaneous query results across headings, frontmatter tags, and meeting transcripts. Review our full-text search architecture.
4. Direct Obsidian Vault Mounting
You can open any existing Obsidian vault folder directly in Oyma. Notes, wikilinks, attachments, and frontmatter remain standard files on disk. The .obsidian configuration folder is completely isolated, ensuring that third-party community plugins never access your audio or memory buffers. See our Obsidian vault compatibility guide.
5. Local Ollama Integration for Structured Summaries
Generate executive summaries, action item checklists with owners, and decision logs using local open-weight language models (Llama 3.1 8B, Mistral, Gemma 2) via the localhost Ollama API. No proprietary cloud tokens or monthly subscription fees are required. See Local AI with Ollama.
Security & Verification Ledger
All 24 automated SEO and security gates passed with zero warnings. The application bundle enforces hardened runtime entitlements, sandbox file restrictions, and network guard verification.
Try private Markdown notes on your Mac
Experience on-device Whisper transcription, instant full-text search, and local Markdown notes in Oyma.