Product

A local runtime for private voice workflows

Izwi is the layer between raw speech models and the applications that need them. Use the desktop app to evaluate, the API to integrate, and Enterprise Runtime when deployment boundaries become operational requirements.

View API docs Download Izwi

Izwi desktop runtime transcription workflow

Product surfaces

One product shape across evaluation, integration, and deployment

Desktop helps teams explore. Server gives applications an API boundary. Enterprise Runtime supports controlled production deployment.

Runtime

The local process that runs speech models and exposes the voice workflow boundary.

OpenAI-compatible API

A familiar /v1 shape for applications that already use modern AI client patterns.

Desktop app

A native evaluation surface for testing models, audio files, and speech workflows.

CLI

A practical control surface for models, services, and local developer workflows.

Enterprise Runtime

Supported packaging, rollout guidance, and operational controls for production environments.

Runtime boundary

Where Izwi fits in the stack

Izwi sits close to the application and data boundary, where speech workflows can be controlled and deployed deliberately.

Deployment targets

developer laptop
Apple Silicon workstation
NVIDIA-backed server
on-prem environment
customer VPC
isolated or offline environment where supported through enterprise packaging

Best at

Private speech workflows with a clear API boundary.

private transcription and diarization

secure dictation and documentation workflows

local voice layers for agent systems

embedded speech capabilities inside regulated software products

API

OpenAI-compatible where it helps

Use a local host, a familiar client shape, and a private runtime boundary.

The integration promise

Applications can keep a familiar integration pattern while the runtime moves closer to the data. Izwi keeps the external shape familiar and changes the deployment boundary underneath.

local base URL for evaluation
same request pattern as hosted AI clients
production packaging when governance matters

Local transcription request

/v1

curl http://localhost:8080/v1/audio/transcriptions \
  -H "Authorization: Bearer not-needed" \
  -F model=qwen3-asr-0.6b \
  -F file=@meeting.wav

Boundaries

Focused on runtime infrastructure

Izwi stays focused on runtime infrastructure, so teams know exactly where it fits.

not a generic cloud speech API

not a collaboration suite

not a catch-all dashboard for every audio task

not a biometrics-first product

See how the runtime fits into your stack

Start with docs if you are integrating. Start with sales if the deployment boundary is already real.

Start with docs Talk to sales