ASR
Speech recognition
Transcribe audio locally with a clean API surface and a deployment path you can keep.
Voice AI runtime
Voice AI that runs inside your environment
Izwi gives teams one local-first runtime for speech recognition, text to speech, diarization, and voice workflows. Start on a laptop, integrate through OpenAI-compatible /v1 endpoints, and move into controlled deployment when the workflow becomes real.
Open source for evaluation. Commercial packaging for production deployment.
One runtime for the core voice stack
Most teams do not need six separate voice tools and six separate integration stories. They need one runtime that can handle the main speech workflows without sending audio to a third-party cloud.
Core runtime
One server process for the workflows teams actually ship.
Instead of stitching together separate vendors for transcription, synthesis, speaker separation, and voice orchestration, Izwi keeps the core stack close to the work and behind one deployment boundary.
Surface area
One API
One runtime shape for local evaluation and controlled deployment.
Data path
No relay
Audio stays inside your environment instead of bouncing through a hosted middle layer.
01
TTS
Text to speech
Generate speech inside your environment instead of routing every request through a hosted vendor.
DIAR
Speaker diarization
Separate speakers in meetings, interviews, and call recordings without moving the audio out.
STACK
Voice workflows
Combine transcription, synthesis, and conversational layers in one local-first stack.
VOICE
Voice cloning
Support custom voice workflows where policy and product fit allow it.
Start local. Deploy with control.
Use Izwi in layers. Explore on desktop, integrate through the runtime API, and move into supported deployment when the workflow matters.
Explore
01Desktop for evaluation
Use the desktop app to test models, prototype workflows, and explore the product locally.
Integrate
02Runtime API for integration
Run the server, point your app at /v1, and integrate without rebuilding your whole stack.
Deploy
03Enterprise Runtime for production
Move to supported builds, deployment guidance, version pinning, observability hooks, and commercial support when the workflow matters.
Built for teams that cannot send audio to the cloud
Izwi is a strong fit when privacy, latency, cost predictability, or deployment boundaries make hosted voice APIs a bad answer.
Best fit
Software vendors and internal platform teams that need a real deployment boundary.
Embed a private voice layer into regulated or customer-controlled workflows. Izwi is strongest when voice is becoming infrastructure rather than a lightweight SaaS add-on. The teams above need control over where the runtime lives, how the integration behaves, and what support path exists when the workflow becomes operational.
regulated workflowsprivate infrastructureoffline-capable deployment
Operational fit
Platform teams
Replace cloud speech dependencies with a runtime you can deploy and govern.
Edge fit
Edge and offline teams
Run voice features where connectivity is weak, expensive, or not allowed.
Familiar to integrate
Izwi keeps the migration story simple. If your team already works with OpenAI-style APIs, the path into a private runtime should feel familiar.
Example
Point the OpenAI client at Izwi
Keep a familiar client, keep the same /v1 shape, and move the runtime closer to the work instead of pushing audio farther away.
- Use the local server for evaluation and prototyping.
- Keep the integration surface stable as deployment changes.
- Use the docs path when you want the fastest first request.
Example: point the OpenAI client at Izwi
/v1from openai import OpenAI
client = OpenAI(
base_url="http://localhost:8080/v1",
api_key="not-needed",
)
transcript = client.audio.transcriptions.create(
model="qwen3-asr-0.6b",
file=open("audio.wav", "rb"),
)The commercial path is simple
Keep the public story aligned to how Izwi actually sells.
Commercial motion
The open path proves fit. The paid path proves operations.
Izwi should feel simple to try and specific to buy. The website needs to make that transition clear instead of treating every visitor like they are already in the same stage.
Phase 01
Evaluate locally
Download the open-source runtime and test the workflow on your own hardware first.
01
Phase 02
Run a paid pilot
Scope one workflow in one target environment and decide what production really needs.
02
Phase 03
Choose the deployment path
Move into Enterprise Runtime, Private Cloud, or OEM when the boundary conditions are real.
03
Private by default. Open by design.
Run the product locally. Inspect how it works. Integrate it through a familiar API. Deploy it where the work already lives.
What the site should make clear
The open-source path is the fast way to evaluate and integrate.
Controlled production deployment belongs to the Enterprise Runtime story.
The runtime, desktop app, and commercial packaging are related, but not identical, product surfaces.
Guardrails
What Izwi is not trying to be
Not another cloud voice API with private deployment added later.
Not a vague AI platform that blurs every surface into one promise.
Not a collaboration-suite story when the real value is deployment control, support, and trust.
Bring voice AI closer to the work
Start locally if you are exploring. Talk to us if you already know the boundary conditions.