Tulpa
Download for Mac Request access
Security & governance

Built so the answer to "where does this go" is short.

This page is the one we'd want if we were the ones vetting a dictation tool: what actually happens to the audio, who can change that, and what we do and don't claim yet.

On-device transcription, always No training on your data, by default Policy enforced at the app layer

Where things stand today

On-device transcription, the local/local-polish/cloud-polish split, and "no training by default" are true right now, for every user. The admin-enforced posture lock described below is rolling out to design-partner teams first — request access to be one of them. We'd rather tell you exactly what's shipped than round up.

The architecture

Five things that make this defensible to your compliance team

01

Transcription happens on the device, under every posture

Speech-to-text runs locally using open, widely-audited speech engines — Whisper-family models and ONNX-based recognizers — rather than a closed black box on someone else's server. Audio is not a network request waiting to happen; it's processed where it was spoken.

02

Only the optional polish pass can ever leave the machine

Tulpa separates transcription (always local) from an optional cleanup pass that fixes filler words and formatting. That pass is the single thing capable of making a network call, and only the resulting text is sent — never raw audio — and only to an endpoint your policy explicitly allows.

03

Posture is enforced, not suggested

Under Local‑only, no external endpoint can be configured by anyone — the option doesn't exist. Under Org‑managed, one endpoint and credential are set centrally and individual overrides are refused, not merely hidden from the UI. Open is available for teams that prefer to let people choose for themselves.

04

Nothing trains on you by default

No audio or transcript is used to evaluate or improve Tulpa's models, or any third-party model, unless your org's chosen posture routes text through an external provider you selected — and that's a deliberate, admin-made choice, not a default anyone opted you into.

05

Deployed fleet-wide, not configured seat by seat

Policy is meant to be pushed once and inherited automatically — new hires get the org's posture on day one, with nothing for an admin to chase down machine by machine. Today Tulpa ships for macOS; Windows and Linux share the same codebase and are on the roadmap.

FAQ

Questions IT and legal actually ask

Does audio ever leave the device?
Under every posture, transcription itself happens on-device. Under Local‑only, nothing leaves at all — no audio, no transcript, no telemetry about what was said. Under Org‑managed or Open with cloud polish enabled, only the transcript text for the optional cleanup pass is sent to the approved endpoint. Raw audio is never transmitted, under any posture.
Can one person turn off the org's policy for themselves?
Not under Local‑only or Org‑managed. The relevant settings are locked, and the app is designed to refuse the override outright rather than simply hide the control — the distinction matters, because a hidden setting can still be reached by anyone who knows where to look.
What platforms does Tulpa run on today?
macOS, today. The codebase is cross-platform by design and Windows and Linux support is on the roadmap — we'd rather ship those well than announce them early.
Is Tulpa open source?
No — Tulpa is a proprietary commercial product. It's built on transparent, widely-used open speech models rather than a closed inference engine, so the core transcription behavior isn't a mystery even though the app itself is closed-source.
Do you train on our audio or text?
No, by default, under any posture we ship. The only way your text could reach a third-party model at all is if your org deliberately configures a cloud-polish endpoint under Org‑managed or Open — and even then, that's about running inference, not about training on your data.
How is this different from a "privacy mode" toggle in other dictation apps?
A per-user toggle depends on every single person remembering to turn it on, and it can usually be turned back off just as easily — there's nothing to audit and nothing stopping drift across a team. Tulpa's posture is set once, by whoever owns the fleet, and enforced at the app layer for everyone under it, rather than left to individual judgment.
Talk to us

Have a longer security questionnaire?

Send it over and we'll walk through the architecture directly — happy to go deeper than a landing page can.

Email hello@tulpa.app Request early access