Privacy Policy

Last updated: September 26, 2026

Saransh is a local-first macOS application operated by Kitto Byte, a California corporation (“Kitto Byte,” “we,” “us”). This page explains the app’s data flow: what is processed on your Mac and which actions contact another service. Local transcription does not mean every feature is offline.

Local processing and storage

• Microphone audio, and optional system audio for Conversations, are transcribed on-device. Conversation capture saves microphone and, when enabled, computer-audio tracks locally so interrupted or failed transcription can be retried. Audio retention is described below. Conversation transcripts are saved locally; Quick Dictation is not saved to the Conversations library.
• Notes are stored in a local SQLite database at ~/Library/Application Support/Saransh/notes.db.
• API keys and license keys are stored in the macOS Keychain under com.kittobyte.saransh.
• If you enable calendar meeting prompts, Saransh reads event titles and start times from macOS Calendar via EventKit on your device to show Start/Skip prompts. Calendar data is not uploaded to Kitto Byte.
• If you enable Capture meeting audio, Saransh mixes microphone and system audio on-device for Conversations. Screen frames are not stored.
• If you enable local MCP for agents, Cursor or Claude Desktop may spawn Saransh --mcp over stdio on this Mac for read-only access to the Conversations you select. This connection stays on your machine; Kitto Byte does not host it. Local transport does not constrain downstream use. Your agent client may send retrieved text to the AI provider configured in that client.
• Apple Intelligence runs on-device when available. Ollama and LM Studio receive text at the server URL you configure: requests stay on your Mac only if that server runs on your Mac. A remote server receives the text.
• Dictation language preferences, spoken editing, vocabulary replacements, and reviewed correction learning are processed and stored locally. Saransh does not automatically learn a correction without your review and approval.

Dictation delivery checks

To deliver dictation, Saransh checks the destination app and available Accessibility field identity. For ChatGPT and Codex desktop windows with unavailable field information, it can use a temporary window identifier and a counter of keyboard, mouse, scrolling, and app-switch activity to check whether the destination has changed. These checks stay on your Mac; they do not record keystroke contents or screen images.

When no field identity is available, Saransh cannot detect every app-driven focus change within the same window. Keep your intended input focused until dictation completes. Recovery text remains available until dismissed or Saransh quits.

Conversation audio, recovery, and deletion

Recordings and recovery metadata are stored under ~/Library/Application Support/Saransh/Recordings/. By default, Saransh deletes conversation audio after the transcript has been successfully saved. In Settings → Conversations → Keep conversation audio, you can instead keep new recordings for 7 days, 30 days, or until you delete them. The choice is saved when a recording starts; changing it does not change the retention of existing recordings. Time-based expiry is measured from recording creation, and expired completed recordings are removed when Saransh next starts.

Failed or interrupted recordings do not expire automatically. They remain available in Meeting Recovery until you delete them or successfully retry transcription and the recording’s retention rule applies. Recovery can use audio already written to disk; it cannot restore samples that were never saved. You can delete a recording’s audio while keeping its transcript and AI notes. Deleting the Conversation also removes its associated audio and chapters. These actions do not delete copies in exports, backups, or other apps, and do not promise secure erasure of underlying storage.

Generated chapters and their source references are stored locally under~/Library/Application Support/Saransh/Chapters/; generated AI notes remain in the notes database. Audio deletion does not delete those text records. The ten-second input test runs locally, saves no Conversation, and attempts to delete its temporary audio after the test. Audio is not included in transcript or AI-note exports or local MCP responses.

What may leave your Mac

• If you generate AI notes, chapters and source-linked findings, or use dictation formatting with a cloud provider (OpenRouter, OpenAI, Anthropic, or Gemini), the request text is sent to that provider. AI notes can include the transcript, title, participant labels, and supplied context; dictation rewriting includes the dictated text and formatting instructions. Chapter generation shows the selected processing destination before you start and sends transcript text, not audio. OpenRouter routes requests to the selected model provider. Those services control their own handling of requests. Provider credentials are stored in the macOS Keychain and used to authenticate requests.
• If you explicitly enable focused-field context for Dictation + rewrite, Saransh may read a limited excerpt around the cursor from the focused editable field. Controls identified as secure text fields by macOS Accessibility are excluded; this is not a general detector of sensitive text. The excerpt is used for that rewrite only, is never saved by Saransh, and is included in the request to your selected AI provider or configured server.
• If you use OpenRouter Connect, Saransh opens OpenRouter in your browser to authorize an API key. AI requests and related account or billing activity are then handled under OpenRouter’s terms and privacy policy.
• License activation sends your license key and an instance name containing your Mac’s name and hardware model to Lemon Squeezy. Validation and deactivation send the license key and activation instance ID.
• Trial registration contacts Kitto Byte’s trial service with a hashed device identifier and the client’s trial-start timestamp so the 14-day trial cannot be reset by reinstalling. Notes, audio, and transcripts are not sent.
• Speech models are downloaded from upstream hosting through the speech-model libraries (WhisperKit or FluidAudio) when needed, including when you select a model that is not cached. Downloads use the network; transcription itself runs on-device.
• If Sparkle auto-update is enabled, the app may checkhttps://saransh.io/download/appcast.xml for available updates and download update packages. These network services receive normal connection information, such as your IP address; the update check does not send your note contents.

Keep content on this Mac

Settings → AI → Keep content on this Mac blocks cloud AI requests and MCP access, including existing agent sessions. It allows available on-device Apple Intelligence and local AI servers at literal loopback addresses such as 127.0.0.1 or [::1]; hostnames and remote addresses are blocked. Enabling it cancels active AI network requests, but cannot recall content already sent. Model downloads, license/trial checks, and updates still use the network. Local servers and other apps can forward content; this control is not a firewall or encryption feature.

Before connecting AI or an agent

For an on-Mac text-processing workflow, use on-device transcription with AI disabled, Apple Intelligence when available, or an AI server running on this Mac. Check any agent client’s provider and data-handling settings separately before enabling MCP. Read-only access prevents database edits; it does not prevent the client from copying or forwarding retrieved text. Exports and text pasted into other apps are handled by those apps and any sync services you use.

Permissions

• Microphone — required for dictation and conversation capture.
• Accessibility — required for global shortcuts and pasting into the focused app.
• Calendar — optional; requested only if you enable meeting prompts in Settings.
• Notifications — optional; used with meeting prompts so you can Start or Skip at meeting time.
• Screen & System Audio Recording — optional; requested only if you enable Capture meeting audio. Saransh uses system audio for Conversations and does not store screen video.

What we do not collect

Saransh does not operate a Kitto Byte-hosted cloud archive of your notes, transcripts, recordings, or calendar events.

Contact

Kitto Byte dba Saransh
support@saransh.io