Speechmark writes your meeting notes — the brief, the decisions, the action items — with whatever AI engine you point it at. The most private options run entirely on your Mac: Apple's on-device models, or a local Ollama model. But not every Mac can run a good local model, and not everyone wants to set one up.
If that's you, Google's Gemini API has a free tier that makes a solid baseline — no credit card, set up in a couple of minutes. This guide walks through getting a key, connecting it to Speechmark, and the one trade-off worth understanding first.
The trade-off, up front
We'd rather you hear this from us than discover it later: on Gemini's free tier, Google may use what you send to improve its models, and human reviewers may read it. That's Google's own policy for free, unpaid use of the API. For meeting transcripts — which can be confidential — that matters.
There are two clean ways around it:
- Turn on billing in your Google account. The moment your project is on the paid tier, Google no longer uses your prompts or responses to train its models. Light meeting use costs cents per month, and you still get a generous free allowance before any charges — billing just changes the data-use terms. This is the option we'd suggest if your meetings are sensitive.
- Use a fully local engine instead. Apple Intelligence or Ollama never send your transcript anywhere. If privacy is the whole point, this is the strongest choice — see Settings → Intelligence in the app.
Either way, one thing stays true: when you bring your own Gemini key, your transcript goes only to Google, under your own account — never to Speechmark. We never see it.
If a free-but-trained-on baseline is a fair trade for you to get started, read on.
Step 1 — Get a free API key from Google AI Studio
- Go to aistudio.google.com/apikey.
- Sign in with any Google account.
- Click Create API key. If it asks which project to use, the default one is fine.
- Copy the key. It starts with
AIza…. Treat it like a password — anyone with it can use your Gemini quota.
That's the whole process. No credit card, no billing setup, no waitlist.
Step 2 — (Recommended) decide on training
This is the moment to make the privacy call from the section above:
- Keeping it free and don't mind the training trade-off? Skip ahead — you're done with Google.
- Want your transcripts kept private? In Google AI Studio, open Billing (or the Google Cloud console for your project) and enable billing on the project your key belongs to. You'll still get the free monthly allowance; you've just opted out of having your data used for training.
Step 3 — Use it in Speechmark
During onboarding: on the Notes engine step, if no local model is detected, pick Use Google Gemini, paste your key, and click Verify. A green check means you're set — Speechmark defaults to the fast, free-tier-friendly gemini-2.5-flash model.
Already past onboarding? Open Settings → Intelligence, choose Google Gemini under Summarization model, and paste your key into the field. Your key is stored in the macOS Keychain on your Mac — not in a config file, and never uploaded to us. To pick a different Gemini model, click Fetch available models.
That's it. Your next meeting will be summarized by Gemini.
What you get on the free tier
Free-tier limits shift over time, but as a rough guide the fast Flash models allow on the order of 1,500 requests per day with a 1-million-token context window — far more than a normal day of meetings needs. Each meeting summary is a small handful of requests, so a typical user never comes close to the cap. The more capable Pro models have much tighter free limits (a few dozen requests a day), so for everyday note-taking, a Flash model is the right default.
You can always see your current limits on the rate-limits page in AI Studio.
Which model should I pick?
gemini-2.5-flash(the default) — fast, cheap, and reliable at the structured JSON Speechmark needs. Best all-round pick.gemini-flash-latest— same idea, but always tracks Google's newest Flash model so it never goes stale.- A Pro model — sharper on long, dense meetings, but the free tier allows very few requests per day. Better suited to a paid (billing-enabled) project.
You can switch any time in Settings → Intelligence — the change applies to your next summary, and never re-sends past meetings.
Prefer to keep everything on your Mac?
Gemini is a convenient baseline, not the only option — and not the most private one. If you'd rather nothing leave your device at all, Speechmark also runs:
- Apple Foundation Models — on-device, zero setup (macOS 26+).
- Ollama — a stronger open model running locally; about five minutes to install.
Both keep your transcripts entirely on your Mac. You can read more about why we built Speechmark around local-first notes on the articles index.