Built for

Client calls, on the record —
your record.

No bot introducing itself to your client. No third-party cloud holding your engagement. A named recording and a speaker-labeled transcript, in your folder, before you write the follow-up.

Download for macOS Apple Silicon · 14-day trial · $29 launch

The awkward moment you are avoiding

A bot named after a note-taking startup joins your client's call. Now your client is thinking about where the recording goes, whether their procurement team approved that vendor, and what happens to it when you finish the engagement. You spend the first two minutes of a paid hour explaining a tool.

Heliotype records from your Mac. Nothing joins, so the call is a call. You still tell them you're recording — that part is not optional and it takes one sentence.

Why it fits how you actually bill

  1. The recording is named after the meeting — AcmeDiscovery-20260728-1000.m4a, from your calendar event, not "Recording 47". Six months later you can still find the call where a scope decision was made.
  2. The transcript sits beside it as plain text with speaker labels and timestamps. It is grep-able, quotable, and it drops straight into a proposal or a scope-change note.
  3. You own the archive. Client files live in a folder you chose — which can be the client's folder, on your own disk, inside whatever backup you already run.
  4. One price, not per seat. $29 once. Transcription bills to your own AssemblyAI key at roughly $0.17–0.23 an audio-hour, which is a rounding error against your rate — and it's a real expense line you can pass through.
The economics matter more than they look. A per-seat note-taker at $17/month — the mid-point of the common tools, billed monthly — is $612 over three years; on an annual plan the cheapest of them is nearer $300. Heliotype is $29 plus what you actually transcribe, either way — the calculator does the arithmetic with your own hours, and shows its assumptions.

The follow-up workflow

Heliotype deliberately does not summarize. You already pay for a model that is better at it than any built-in template, and it is the model you have taught your own context to. Drag the transcript into Claude or ChatGPT and ask the question you actually have — "what did I commit to, and by when?", "draft the recap in my voice", "list every objection they raised". Clean input, your model, your judgment.

Confidentiality, stated plainly

Recordings and transcripts stay on your Mac. When you transcribe, the extracted audio goes to your AssemblyAI account under your key — you hold that relationship and their terms directly, and nothing routes through a server of ours, because there isn't one. No account, no telemetry, no analytics. If a client asks what tooling touches their call, the answer is short and checkable: the privacy page.

If a client contract forbids recording, this changes nothing about that. Ask first, as you already do.

Also worth reading: what it can't do — no summaries, no live captions, no Windows — and how recording without a bot works. Related: when your client's IT blocks bots, interview work, and the full FAQ.