V 0.6.1 BETA

Intelligence without
the cloud.

A local-first, zero-latency voice assistant engine built with Rust and Tauri. Your audio never leaves localhost.

$ gibberish --status
[INFO] Audio Bus: Active (Zero-Copy Arc<[f32]>)
[INFO] Latency: 45ms | RTF: 0.12
[INFO] Model: Sherpa Zipformer (Streaming)
[INFO] VAD: Silero (Active)
[INFO] SmartTurn: Enabled
[SUCCESS] Listening on CoreAudio Device...

$ _

// Core Features

01. Zero Latency

Words appear char-by-char as you speak. Powered by a Rust-native audio bus that bypasses the JS bridge entirely.

02. 100% Private

Runs offline. No API keys, no servers, no telemetry. Your voice data is processed in RAM and dropped.

03. Smart Turn

Semantic endpoint detection. It knows when you've finished a sentence using neural analysis, not just silence detection.

04. Multilingual

Native support for English and Catalan (NeMo CTC). Plus Whisper Small integration for 99+ other languages.

05. Agentic

Local LLM (FunctionGemma) analyzes your speech to trigger desktop tools. "Search Wikipedia for Rust" just works.

// The Pipeline

A high-performance modular monolith. Separation of concerns between domain logic and application glue.

Microphone cpal / CoreAudio
Recorder Native Plugin
Audio Bus Arc<[f32]>
STT Worker std::thread
Inference Sherpa ONNX