Privacy policy

Your meetings stay on your machine.

Effective May 14, 2026

Who runs Speechmark

Speechmark is built and operated by ChakraBit Solutions LLP, a limited liability partnership registered in India (chakrabit.com). For privacy questions, write to [email protected]. Under data-protection laws that use the term, ChakraBit Solutions LLP is the data controller for the small amount of personal data described below.

The short version

  • Audio never leaves your Mac. Recording files are written to a local folder and stay there until you delete them.
  • Transcripts only leave your Mac if you choose a cloud model for summaries (Anthropic Claude or OpenAI). The default — Apple Foundation Models — and the Ollama option both run fully on-device.
  • There is no Speechmark user account. Anonymous usage statistics are opt-in and off by default — if you turn them on, the app sends counts (how many meetings, total minutes, notes generated) and which engine you picked, never your meeting content. Everything Speechmark-operated is described below.
  • You can delete every recording, transcript, and note yourself, in the app, in one click — or by removing one folder on disk.

What Speechmark captures

When you start a recording, Speechmark captures two audio streams on your Mac: your microphone (so it can hear you) and system audio (so it can hear everyone else through your speakers or headphones). It does not join your meeting as a participant, and there is no bot in the attendee list.

Both streams are written to local files under ~/Library/Application Support/Speechmark. Transcripts, notes, and meeting metadata sit alongside them in the same folder. Nothing in that folder is uploaded anywhere by us.

What leaves your Mac, by provider

Speechmark turns recordings into a clean note using a language model. You choose which model. That choice is the only thing that determines whether any data leaves your Mac.

In every case below, the audio file itself is never uploaded. Only the transcript text — and any notes you add — are sent.

Apple Foundation Models
DEFAULT · ON-DEVICE
Nothing leaves your Mac. The summary is generated entirely on-device by Apple Intelligence. Requires a Mac that supports Apple Intelligence.
Ollama
LOCAL · OPT-IN
The transcript is sent to the Ollama server you run yourself, on this Mac. No network call leaves the machine. No API key. You pick the model (e.g. Llama 3, Mistral).
POST http://127.0.0.1:11434/api/chat
About Ollama
Anthropic Claude
CLOUD · OPT-IN · BYOK
The full transcript text and any optional notes you wrote are sent to Anthropic so the model can produce a summary. The request uses an API key you provide.
POST https://api.anthropic.com/v1/messages
Anthropic data-handling policy
OpenAI
CLOUD · OPT-IN · BYOK
The full transcript text and any optional notes you wrote are sent to OpenAI so the model can produce a summary. The request uses an API key you provide.
POST https://api.openai.com/v1/chat/completions
OpenAI API data policy

How long things are kept

On your Mac, recordings and transcripts are kept until you delete them. There is no automatic expiry. If you reinstall macOS or move to a new machine without copying the folder over, your data is gone.

For cloud models, retention is governed by the provider you chose. We link to their policies in the table above. You should read those if you are recording anything sensitive.

Deleting your data

Every meeting in Speechmark has a delete action that removes the recording files, the transcript, and the meeting record from disk. There is no soft-delete, no trash, no recovery — when you delete a meeting, the files are gone.

If you want to nuke everything at once, quit Speechmark and remove ~/Library/Application Support/Speechmark. We never receive a copy of your meetings, so there is nothing to delete on our side.

To delete the personal data we do hold (a beta-application row, or your purchase email and license-key record — see the next section), email [email protected] from the address on file and we will remove it within 30 days. Note that deleting the license record disables further validation of that key.

What talks to Speechmark's own servers

Speechmark has one small backend, a Cloudflare Worker at api.speechmark.co. It does a handful of things, and only these things:

  • Beta application form. When you fill out the form on speechmark.co, we store the fields you submitted (email, name if given, role, meetings/week, macOS version, current tool, free-text reason) so we can review and email you back. Stored in Cloudflare Workers KV, retained until we close the beta cohort or you ask us to delete it.
  • License validation. When the macOS app first activates a license key (or re-validates), it sends the key to api.speechmark.co. We log the timestamp and a hash of the key to track activation count. We do not see, send, or store anything about your meetings.
  • Purchase webhook. When you buy Speechmark, the payment processor (Lemon Squeezy) posts your name, email, and order ID to api.speechmark.co so we can issue your license key. We don't see card details — those stay with Lemon Squeezy. See Lemon Squeezy's privacy policy.
  • Anonymous usage statistics (opt-in, off by default). We ask once during setup, and there is a toggle in Settings. If — and only if — you turn it on, roughly once a day the app sends a random install ID (not tied to your name, email, license, or machine), the app version and macOS major version, running totals of how many meetings you've recorded, total minutes, and notes generated, and which engine you chose (transcription backend, LLM provider, whether speaker labels are on, trial vs. licensed). It never sends meeting audio, transcripts, note text, titles, file names, or contacts. Turn it off and the app stops sending it.
  • In-app feedback. If you choose to send feedback from the prompt inside the app, we receive the rating you picked, any comment you typed, the build ID, and a hash of your license key (not the key itself). This only ever happens when you tap “Send feedback” yourself.

That is the entire list. There is no crash-reporting service and no sync server. Usage statistics are opt-in and off by default — with them off, the Mac app sends nothing but license validation. On any setting, it never sends your meeting audio, transcripts, note text, titles, or contacts.

Your rights over the data we hold

If you live in the EU/EEA, the UK, California, or another jurisdiction with data-protection laws (GDPR, UK GDPR, CCPA/CPRA, India's DPDP Act), you have rights over the small amount of personal data Speechmark holds about you (the items in the section above): access, correction, deletion, portability, and the right to object to processing or withdraw consent. To exercise any of these, email [email protected] from the address on file. We respond within 30 days.

Our lawful basis for the data we do collect is contract performance (we need your email to send you a license key) and your consent (you chose to apply for the beta). We do not use your data for marketing and we do not sell or share it.

What we don't do

  • No telemetry
  • No analytics or pixel tracking
  • No crash reporting service
  • No Speechmark account
  • No cloud sync server
  • No ads, ever
  • No selling or sharing data
  • No training models on your meetings

Your responsibility — recording other people

Speechmark records audio your Mac can hear. Whether you should record a given conversation, and whether you are legally required to tell the other participants, depends on where you are, where they are, and what kind of conversation it is. Many places require all-party consent. Some require only single-party consent. A few places make recording calls illegal outright.

macOS shows a microphone indicator in the menu bar while Speechmark is recording. That's a system-level cue for you — it is not the same as informing the room. Telling people you're taking notes with an AI assistant is your responsibility, and we recommend doing it even where the law doesn't require it. It's also good manners.

We protect your data on your Mac. We can't get consent from the other people in your meeting for you. That part is on you.

Helpful background: telephone call recording laws by jurisdiction. Not legal advice.

Your API keys

If you enable a cloud model, you supply your own API key for that provider. Those keys are currently stored in your Mac's UserDefaults database; we are moving them to the macOS Keychain in an upcoming release. Either way, your keys are used only to authenticate the calls described in the provider table above. They are never sent to us — we don't have a server to receive them.

Updates to this policy

When we change how Speechmark handles your data, we'll update this page and bump the effective date at the top. Material changes — new providers, new endpoints, new categories of data leaving your Mac — will be called out in the release notes for the version that introduces them.

Contact

Questions or concerns about privacy? [email protected].