v0.2.0 · Native MLX audio + local Ollama text

Ship voice AI that can change its mind.

Vona is the Rust runtime layer for voice-native products: fast, provider-neutral speech-to-speech infrastructure with native MLX audio adapters, local Ollama text generation, deterministic interruption tests, and transport-ready session orchestration.

Start building View source
vona session trace
transport MLX Ollama realtime
01 audio.frame.in 20ms
02 mlx.whisper.transcribe metal
03 interruption.detected barge-in
04 ollama.generate.stream loopback
05 mlx.qwen3.synthesize native
16 Workspace crates
6 Provider surfaces
4 Local engines
MIT Open source
Runtime substrate

The layer between a demo and a real voice product.

Vona owns the reusable runtime boundary: sessions, frames, tools, generation, fallback decisions, and backend-neutral contracts. Your product owns the experience, policy, and deployment.

Speech-to-speech contracts

Step-oriented and event-stream voice interfaces give hosted realtime APIs, local STS models, native MLX speech engines, and sidecar transports one coherent Rust boundary.

AudioInputFrame TextGenerator BackendStep RealtimeVoiceOutput

Interruption-aware sessions

Measure time-to-first-audio, tool calls, interruption behavior, output frames after barge-in, and fallback decisions in tests before users discover the edge cases.

Native MLX speech

Keep Whisper STT and Qwen3 TTS in native Rust crates while targeting Apple Silicon Metal through mlx-rs and shared speech model loaders.

Tool context without glue sprawl

Skill registries, schema validation, audit events, and external context injection let voice systems coordinate with application tools while keeping runtime policy explicit.

SkillRegistry ExternalContextEvent OllamaTextEngine FallbackPolicy
Backend portability

Local when you need control. Cloud when you need reach.

Vona is designed for mixed reality: one app can explore local Moshi and Seamless-style paths, native MLX speech, local Ollama text reasoning, cloud realtime protocols, STT/TTS cascades, and sidecar deployments without hard-coding its future to a single provider.

OpenAI Realtimeevent-stream mapping
Gemini Liverealtime protocol surface
Azure Voice Livevoice and speech helpers
ElevenLabsstreaming TTS surface
DeepgramFlux, listen, Aura helpers
Moshi / Seamlesslocal STS families
vona-mlxWhisper STT and Qwen3 TTS
vona-ollamalocal HTTP text generation
Release confidence

Test the voice moments that normally feel impossible to pin down.

Vona’s deterministic harnesses make the slippery parts of speech runtime behavior observable: event order, first audio, tool calls, interruption cleanup, local model behavior, fallback paths, and transport latency.

release gate 
$ cargo run -p vona-test-harness --example mock_session --locked
session_id=mock-session-1
close_reason=BackendFinished
metrics time_to_first_audio_ms=Some(0) tool_calls=1 interruptions=1 fallback_count=0
output_frames_after_interruption=0
injected_events=1

$ bash scripts/release_gate.sh
[release-gate] Release gate PASSED
Start with the facade

One dependency. Opt-in surfaces.

Use the umbrella crate for applications, then enable only the adapters you actually ship. Drop lower when you need direct crate boundaries for Ollama text, MLX Whisper STT, or MLX Qwen3 TTS.

Read the README Changelog
Cargo.toml 
[dependencies]
vona = { version = "0.1.0", features = [
  "ollama",
  "mlx-whisper-native",
  "mlx-qwen3-tts-native",
  "openai-realtime",
] }