PRODUCTS

Building blocks for on-device voice agents

A local voice stack, C++ agents SDK, and models for building speech-driven systems that run on-device.

Voice AI

Voice AI Stack

Local speech-to-text, text-to-speech, and voice activity detection for robots, vehicles, and embedded devices.

View Voice AI

Voice Stack

  • Text-to-speech with multiple expressive voices
  • Real-time speech-to-text with webhook events
  • Voice activity detection
  • Streaming LLM-to-speech pipeline
  • HTTP API with token-based auth

Best For

  • On-device voice assistants
  • Offline speech processing
  • Privacy-first voice interfaces
  • Push-to-talk and continuous listening modes

Agent Harness

Agent Harness

C++ SDK for building local agents that reason, plan, call tools, and orchestrate multi-step workflows.

View Agent Harness

SDK Features

  • Multi-modal autonomous agents
  • Extensible tools
  • High-performance C++ runtime
  • Workflow patterns
  • Evaluation and optimization tools

Agent Capabilities

  • In-vehicle voice assistants
  • Robotics task control
  • Manufacturing operations assistants
  • Offline emergency services assistants

Model Catalog

Model Catalog

Open-weight language models from 1B to 35B parameters, curated and optimized for on-device deployment.

View Model Catalog

Size Tiers

  • Ultra-Light 1-4B for phones, IoT, and embedded devices
  • Small 4-14B for laptops, Jetson, and single GPU systems
  • Edge Server 14-35B for workstations and edge servers

Selected For

  • Reliable tool calling and function execution
  • Reasoning workflows
  • Multimodal capabilities across vision and audio
  • Open weights with Hugging Face availability