Saransh

Changelog

What’s new

Release notes for Saransh for Mac. Subscribe via the app or check back after updates.

July 15, 2026

v1.4.0 — Meeting audio, context modes, vocabulary, local AI, and MCP

  • Opt-in Capture meeting audio for Conversations: mix microphone with system audio via ScreenCaptureKit (audio only; Quick Dictation stays mic-only).
  • App-aware Quick Dictation context modes: Polish, Email, Notes, To-do, and Agent prompt — auto-selected by frontmost app bundle ID, with custom profiles.
  • Vocabulary & replacements in Settings → Speech: Whisper recognition hints plus post-ASR find→replace for dictation and conversations.
  • Speaker edit tools: rename, merge, and reassign turns after diarization.
  • Optional auto-stop when a calendar meeting ends, and optional auto-export Conversations as Markdown to a chosen folder.
  • Local AI providers for formatting and Enhance: Ollama, LM Studio, and Apple Intelligence alongside OpenAI, Anthropic, and Gemini.
  • Opt-in local MCP notes API for Cursor and Claude Desktop (`Saransh --mcp`) — read-only list/search/get over stdio, gated from Settings → Agents.
  • Import VTT or plain-text transcripts into a conversation note.

July 12, 2026

v1.3.0 — Calendar meeting prompts

  • Opt-in calendar meeting prompts via macOS Calendar (EventKit), including accounts already synced in Calendar.app.
  • On-screen Start/Skip prompt at meeting start — recording never begins until you tap Start.
  • Meeting notes can take the calendar event title when you start from the prompt.
  • Single-instance guard so notification actions do not spawn a second Saransh process.
  • Hotkey stop path hardened so ending a calendar-started conversation does not accidentally start a new one.

July 2026

v1.2.0 — Smart formatting for Quick Dictation

  • Smart formatting modes for Quick Dictation: Off, Polish, or Engineered Prompt (later evolved into app-aware context modes in 1.4.0).
  • Polish: optional punctuation and grammar cleanup before paste (BYOK).
  • Engineered Prompt: turn dictated thoughts into a structured agent prompt for Cursor and similar tools, with editable instructions (BYOK).

June 29, 2026

v1.1.0 — Whisper model tiers

  • Three on-device transcription tiers: Fast (base.en), Balanced (large-v3 turbo), and Best Accuracy (large-v3 compressed).

June 27, 2026

v1.0.0 — Initial release

  • Menu bar notepad with global Quick Dictation (⌃⌥X) and Conversation capture (⌃⌥R).
  • Configurable shortcuts with hold or toggle capture modes.
  • On-device Whisper transcription with one-time model download.
  • Optional on-device speaker diarization with inline speaker rename.
  • Optional dictation polish before paste (requires your API key).
  • Full-text search across titles, transcripts, and AI notes (SQLite FTS5).
  • AI note enhancement with bring-your-own-key support for OpenAI, Anthropic, and Gemini.
  • Built-in templates: General Summary, Meeting Recap, 1:1 Sync, Standup, and Action Items — plus custom templates.
  • Multiple AI notes per conversation with markdown preview and interactive task checklists.
  • Export transcripts or AI notes as Markdown (`.md`).
  • 14-day full trial and lifetime licensing via Lemon Squeezy.
  • Local SQLite storage and Keychain-backed secrets.