What stays local
All transcription — file, dictation, and live captions — runs on hardware controlled by the user. Audio is processed by local .oasr model packs from the signed catalog. No audio data is sent to any server during normal operation.
There is no telemetry or usage analytics in the CLI, server, or desktop app. No account or sign-in is required.
Network exceptions
OpenASR contacts the network in a small number of explicit scenarios:
- Explicit model downloads approved by the user.
- Desktop update-manifest checks.
- An optional user-supplied Hugging Face token sent only to huggingface.co for eligible model downloads.
- Optional Remote Compute, which is off by default.
How Remote Compute works
OpenASR does not provide or operate a central cloud transcription backend. Remote Compute sends audio only to a user-controlled device running OpenASR Server that you have explicitly paired using TLS, trust-on-first-use certificate pinning, and a safety code.
You must deliberately enable Remote Compute and pair the target server yourself. It is never activated silently.
CLI offline mode
The CLI --offline option forbids network access. If a required model is not already present locally, the CLI fails rather than downloading it.