Saransh

Local-first · macOS 14+

Speak. Capture. Structure.

Local-first menu bar notepad for macOS. Dictate into any app with app-aware formatting, capture conversations (optional meeting audio) with on-device Whisper, enhance with cloud or local AI, and let agents read notes over opt-in MCP.

14-day full trial · $59 lifetime · One Mac at a time · Microphone & Accessibility required; Calendar and meeting audio optional

Saransh is a product of Kitto Byte, a California corporation.

Saransh — Speak. Capture. Structure.
⌃⌥XEmailQuick Dictation

Listening…

Hi team — shipping the onboarding polish Friday. Can you review the privacy FAQ?

Pasted into Mail · not saved to library

Team sync

Conversation · On-device Whisper

REC

Speaker 1

Let’s lock the launch timeline this week.

Speaker 2

I’ll own the privacy FAQ draft by Thursday.

Team sync recap

AI Note · Meeting Recap

Summary

Discussed launch timeline, onboarding flow, and local-first privacy positioning.

  • Ship website redesign
  • Draft privacy FAQ answers
⌃⌥X Dictate⌃⌥R RecordLocal MCP

Features

See it in action

Dictate with app-aware formatting, capture both sides of a call when you opt in, then enhance with cloud or local AI — each pillar shows the workflow and the details around it.

1

Dictate

Speak into any app. Format for the moment.

Hold ⌃⌥X anywhere. Saransh transcribes locally, pastes into the focused field, and never saves to your library. Overlay shows the active context mode — Polish, Email, Notes, To-do, Agent prompt, or a custom profile bound to a bundle ID.

  • Global shortcut with hold-to-talk flow
  • App-aware modes; unbound apps use Polish
  • Never saved to your Conversations library
Quick Dictation · any app
Hold ⌃⌥X
EmailNotesAgent prompt
App-aware mode

#launch-channel

Shipping onboarding polish Friday — can you review the privacy FAQ before we cut the release?

Transcribed locally · never saved

Context modes

Bound modes to apps

Mail → Email, Cursor → Agent prompt. Unbound apps use Polish.

Mail.appEmail
SlackPolish
CursorAgent prompt

Vocabulary

Teach names & jargon

Hints for Whisper, then deterministic find→replace after transcription.

Heard

“ping jon do on thejira ticket

Replaced

“ping John Doe on theJIRA ticket

Conversations · optional meeting audio

Calendar

Team sync started

Record this conversation? Recording never begins until you tap Start.

StartSkip

No meeting bots · opt-in EventKit

RecordingMic + system audio02:14

On-device Whisper · stays on your Mac

Live transcript

Speaker labels

Speaker 1

Let’s lock the launch timeline this week.

Speaker 2

I’ll own the privacy FAQ draft by Thursday.

Speaker 1

Great — rename speakers anytime after.

2

Capture

Conversations with optional meeting audio.

Press ⌃⌥R, or start from an opt-in calendar Start/Skip prompt. Mix microphone with system audio so remote participants are included — audio only, no screen video stored. Speakers are labeled on-device; rename, merge, or reassign turns anytime.

  • No meeting bots — you tap Start
  • On-device Whisper (Fast / Balanced / Best)
  • Each session becomes a searchable note

Meeting audio

Mic + system audio

Opt in so remote participants are included. Audio only — no screen video.

Microphone

System

Mixed into one conversation

On-device Whisper

Pick your accuracy

Fast, Balanced, or Best — model downloads once, speech stays on your Mac.

Fast
Balanced
Best Accuracy

Speakers

Rename & merge

Label Speaker 1 / 2 on-device, then rename or reassign turns.

Speaker 1Alex

Let’s lock the launch timeline.

Speaker 2Sam

I’ll draft the privacy FAQ.

3

Enhance

Structure notes. Hand them to agents.

Run built-in templates — General Summary, Meeting Recap, 1:1 Sync, Standup, Action Items — with OpenAI, Anthropic, Gemini (BYOK), or keep it local with Ollama, LM Studio, or Apple Intelligence. Preview markdown with interactive task checklists. Opt in to local MCP so Cursor and Claude Desktop can read Conversations — read-only, nothing uploaded.

  • Cloud BYOK or fully local AI
  • Export `.md` or auto-export to a folder
  • MCP never exposes Quick Dictation
AI Note · Meeting Recap
TemplateMeeting Recap

Raw transcript

Speaker 1: Let’s lock the launch timeline this week. Speaker 2: I’ll own the privacy FAQ draft by Thursday…

Enhancing with cloud or local AI…

Team sync recap

Markdown preview · interactive checklists

Summary

Discussed launch timeline, onboarding flow, and local-first privacy positioning.

  • Ship website redesign
  • Draft privacy FAQ answers

Local MCP

agent → list_conversations()
agent → read(“Team sync”)

Cursor & Claude Desktop · read-only · on this Mac

Opt in from Settings → Agents · Quick Dictation never exposed

AI providers

Cloud BYOK or local

OpenAI, Anthropic, Gemini — or Ollama, LM Studio, Apple Intelligence.

OpenAIAnthropicGeminiOllamaLM StudioApple Intelligence

Your keys stay in Keychain. Local models never leave the Mac.

Templates

Structure in one click

Meeting Recap, Standup, Action Items — plus custom prompts.

Running

Meeting Recap1:1 SyncAction Items

Decisions extracted

Next steps listed

Search & export

Find it. Ship the `.md`.

FTS5 across titles, transcripts, and AI notes. Export or auto-export to a folder.

privacy FAQ
Team sync recap · matched in AI note

/Notes/Saransh/

team-sync-recap.md

transcript.md

Privacy

Local-first by design

Your voice and notes stay on your Mac unless you explicitly send text to a cloud AI provider you choose — or keep enhance fully local.

100%

On-device transcription

Whisper runs locally on your Mac

0

Kitto Byte cloud archive

We do not host your notes or recordings

Local+

Optional AI enhance

Cloud BYOK or Ollama / LM Studio / Apple Intelligence when available

Keychain

Secrets storage

License and API keys stay in macOS Keychain

Pricing

Lifetime license

$59 one time

  • 14-day full trial on first launch (device-bound)
  • No usage caps on dictation or conversation capture
  • One Mac at a time — deactivate to move
  • App-aware context modes for Quick Dictation
  • Vocabulary hints & replacements
  • Opt-in calendar Start/Skip prompts
  • Opt-in meeting audio (mic + system audio)
  • AI enhance via cloud BYOK or local models
  • Opt-in local MCP for Cursor & Claude Desktop
  • Markdown export (`.md`) and Sparkle updates for v1.x
Buy on Lemon Squeezy

No subscription. Activate with the key from your purchase email. One Mac at a time.

Setup

Getting started

Up and running in minutes. Most of the wait is the one-time speech model download.

  1. 1Download the DMG and drag Saransh to Applications.
  2. 2Grant Microphone and Accessibility permissions during onboarding.
  3. 3Wait for the speech model download (first launch only).
  4. 4Optionally enable calendar meeting prompts or Capture meeting audio in Settings → Conversations.
  5. 5Optionally bind apps to context modes (Email, Agent prompt, and more) in Settings → Quick Dictation.
  6. 6Add a cloud API key or choose Ollama / LM Studio / Apple Intelligence (when available) if you want AI notes or formatting.
  7. 7Optionally add vocabulary & replacements in Settings → Speech, and enable local MCP in Settings → Agents.
  8. 8Activate your license key from the purchase email before the trial ends (Settings → License). One Mac at a time — deactivate to move.

FAQ

Common questions

Quick answers before you download. More detail is in our privacy and terms pages.

What are the system requirements?

macOS 14 or later. Saransh needs Microphone and Accessibility permissions. Calendar and Notifications are optional (meeting prompts). Screen & System Audio Recording is optional (Capture meeting audio). Internet is required on first launch to download the on-device Whisper speech model. An AI provider is optional and only needed for AI notes or dictation formatting — cloud BYOK (OpenAI, Anthropic, Gemini) or local (Ollama, LM Studio, or Apple Intelligence when available on your Mac).

What permissions does Saransh need?

Microphone for dictation and conversation capture, and Accessibility to paste transcribed text into the app you are using. Calendar and Notifications are requested only if you enable meeting prompts. Screen & System Audio Recording is requested only if you enable Capture meeting audio — Saransh uses audio only and does not store screen video.

How do calendar meeting prompts work?

The feature is off by default. When enabled, Saransh reads timed events from macOS Calendar (including Google or Outlook accounts already synced there). At meeting start it shows an on-screen Start/Skip prompt — recording never begins until you tap Start. You can optionally auto-stop when the calendar event ends. Saransh does not join calls as a bot.

Can I fix names and jargon in transcripts?

Yes. In Settings → Speech, add vocabulary words as Whisper recognition hints and set replacements (e.g. misspelled names → correct spelling). Replacements apply after transcription for both Quick Dictation and Conversations.

Can Saransh capture both sides of a call?

Yes, when you enable Capture meeting audio in Settings → Conversations. Conversations then mix your microphone with system audio so remote participants are included. This requires Screen & System Audio Recording permission. Quick Dictation stays microphone-only. You are responsible for complying with local consent and recording laws.

What are context modes?

Quick Dictation formats what you said using instruction profiles bound to apps (by bundle ID). Built-ins include Polish (default for unbound apps), Email, Notes, To-do, and Agent prompt. You can edit instructions and add custom profiles in Settings → Quick Dictation. Formatting needs a configured AI provider; without one, Saransh pastes the raw transcript.

Which Whisper model should I use?

Saransh offers three on-device tiers in Settings: Fast (~140 MB) for the quickest download and lowest latency, Balanced (~630 MB) as the recommended speed/accuracy tradeoff, and Best Accuracy (~626 MB) for the highest transcription quality. Each tier downloads once; switch anytime in Settings or the menu bar.

Does my audio leave my Mac?

Transcription runs on-device with Whisper. Audio and transcripts stay in your local SQLite database. If you use a cloud AI provider for enhance or dictation formatting, transcript text is sent to that provider with your API key. Ollama, LM Studio, and Apple Intelligence (when available) keep that step local.

Do I need an API key?

No. Dictation and conversation capture work without one. Add an OpenAI, Anthropic, or Gemini key only if you want cloud AI formatting or notes. Or choose Ollama, LM Studio, or Apple Intelligence (when available on your Mac) in Settings for local enhance and formatting with no cloud key.

What is local MCP?

An opt-in, read-only connector for Cursor and Claude Desktop. Enable Allow local MCP for agents in Settings → Agents, copy the config into your client, and agents can list, search, and read Conversations over stdio on your Mac (`Saransh --mcp`). Nothing is uploaded. Quick Dictation is never exposed.

How does the trial work?

You get a 14-day full trial on first launch with no feature limits. The trial is bound to your Mac, so reinstalling does not reset it. After the trial, capture and enhance require a lifetime license activated from your purchase email. Existing notes remain readable.

How do I activate my license?

Open Settings → License in Saransh and paste the key from your Lemon Squeezy purchase email. Activation validates with Lemon Squeezy over the network. Each lifetime license activates on one Mac at a time — deactivate in Settings → License before moving to another computer.

What can I export?

Markdown only (`.md`) for a raw transcript or any AI note. You can also optionally auto-export Conversations as `.md` files into a folder you choose (for example an Obsidian vault). Auto-export is one-way and overwrites the same filename; there is no two-way sync.

What is your refund policy?

Saransh includes a 14-day full trial before you buy. After purchase, lifetime licenses are non-refundable except for billing errors, activation issues we cannot resolve, or where required by law. See our Refund Policy for details.

Ready to speak your notes?

Download Saransh, try every feature for 14 days, and keep your transcripts local on your Mac.