Local ASR toolkit

Beyond a single Whisper binary

If you already run Whisper-class models locally—whisper.cpp or similar—you may want a fuller local stack: a signed model catalog, a Desktop app on macOS and Windows, and a Rust CLI with a local HTTP API. OpenASR is that local-first alternative. It is not a hosted Whisper API and does not claim to be the best engine on every benchmark.

Open core
Apache-2.0 Rust CLI and local HTTP API
Desktop
Closed-source apps for macOS and Windows only
Models
Signed catalog with explicit install; no silent auto-download on normal STT
Extras vs bare CLI engines
MOSS separation, Voice ID on Desktop; --diarize and /v1/speakers in core

Who this is for

Engineers and power users who like local Whisper-class quality but need packaging around it: consistent model distribution, a local server API, desktop UX, and optional speaker workflows.

OpenASR keeps the recognition path on your hardware. There is no account gate and no central cloud transcription backend for standard use.

Catalog, not ad-hoc weights

Instead of hunting unmarked weight files, you pull from a signed catalog. That makes integrity checks and repeatable installs part of the default path.

Normal transcription does not silently auto-download. CLI --offline fails closed when the local tree is incomplete—useful in CI images and locked laptops.

  • Signed catalog for model distribution
  • Local HTTP API for app and script integration
  • Desktop for day-to-day capture and review on Mac/Windows
  • Diarization paths in both Desktop and open core

Desktop when you leave the terminal

whisper.cpp-adjacent workflows are often terminal-first. OpenASR Desktop adds a product UI for transcription, MOSS multi-speaker separation, and Voice ID naming on supporting models—still local by default.

There is no native Linux desktop app; keep using CLI/API or containers on Linux.

API-shaped local integration

The local HTTP API, including speaker endpoints under /v1/speakers with a WeSpeaker capability pack, is aimed at tools that should call localhost rather than a vendor SaaS URL.

Docker Hub images (CPU multi-arch and CUDA) exist for server-style deploys; they are complementary, not the focus of the Desktop experience.

Frequently asked questions

Is OpenASR a hosted Whisper API?

No. It is local-first software: open-core CLI/API plus Desktop apps. Transcription is not routed through a central OpenASR cloud backend.

How is this different from running whisper.cpp alone?

You still stay local, but OpenASR adds a signed model catalog, Desktop on macOS/Windows, a local HTTP API, and first-class speaker separation/enrollment paths. It is a broader toolkit, not only a single decoder binary.

Do you publish WER leaderboards against Whisper?

No. OpenASR does not invent benchmark or WER numbers on these pages and does not claim unqualified “best” status. Evaluate on your audio and hardware.

Can I automate OpenASR like a local service?

Yes. Use the Rust CLI and local HTTP API, including diarization flags and the /v1/speakers API. Containers are available when you want an image-based deploy.

Is the whole stack open source?

The core CLI and local HTTP API are Apache-2.0. Desktop apps for macOS and Windows are closed-source product builds on top of that core.

Offline speech to text Mac and Windows transcription that stays on-device by default. Models Signed catalog across Whisper and other families. Server API OpenAI-compatible localhost endpoints. Download Desktop apps and prebuilt CLI releases.