Offline STT

Speech to text that works without the cloud

OpenASR is local-first speech-to-text: transcription runs on your machine with models you install from a signed catalog. Desktop covers macOS and Windows; the open core adds a Rust CLI and local HTTP API. There is no account, no telemetry, and no central cloud transcription backend—and CLI --offline fails closed instead of reaching for the network.

Desktop platforms
macOS and Windows Desktop apps; no native Linux desktop app
Network posture
No account, no telemetry, no central cloud STT backend
Model source
Signed catalog; no silent auto-download on normal transcription
Strict offline
CLI --offline fails closed when local requirements are not met

What “offline” means here

Offline speech to text means the recognition path does not depend on a hosted ASR API. Audio and transcripts stay with the processes on your computer, subject to the models and packs you have already installed.

OpenASR makes that default explicit: local-first product design, signed model catalog, and a CLI flag that refuses to proceed when offline guarantees cannot be kept.

Desktop on Mac and Windows

Closed-source OpenASR Desktop apps provide a local transcription experience on macOS and Windows, including meeting-oriented features such as MOSS speaker separation and Voice ID on supporting models.

You do not create an account to transcribe. Ordinary use does not send audio to a central OpenASR cloud backend.

  • Install Desktop on macOS or Windows
  • Install models you choose from the signed catalog
  • Transcribe locally for notes, meetings, and interviews
  • Use Voice ID and MOSS when you need speaker structure

CLI and local API for locked-down environments

The Apache-2.0 open core is a Rust CLI plus a local HTTP API. That surface fits air-gapped labs, regulated laptops, and automation where a GUI is optional.

Pass --offline when you need fail-closed behavior: if a required model or asset is not available locally, the command errors instead of downloading or calling out.

Models without surprise downloads

Models are published through a signed catalog so integrity is checkable. Normal transcription does not silently auto-download models in the background; you remain in control of what lands on disk and when.

Frequently asked questions

Does OpenASR require internet to transcribe?

Not for ordinary local transcription once the models you need are installed. The product is local-first and has no central cloud transcription backend. CLI --offline fails closed if local requirements are missing.

Is there telemetry or a mandatory account?

No. OpenASR does not require an account and does not use telemetry as part of the described product posture.

Can I run offline speech to text on Linux?

Yes via the open-core CLI, local HTTP API, or container images. There is no native Linux desktop app.

Will models download by themselves during transcription?

Normal transcription is designed not to silently auto-download models. You install from the signed catalog deliberately; --offline refuses network fallback.

Which desktop operating systems are supported?

OpenASR Desktop targets macOS and Windows. Linux users should use CLI, API, or Docker images rather than a native desktop shell.

What is OpenASR Local-first architecture, open core, and desktop apps. Does it upload audio? Default local path and the few explicit network exceptions. Local Whisper alternative Desktop app, signed catalog, and local API without a cloud backend. Download macOS and Windows desktop, plus CLI on Linux.