Own your expression, find your freedom
VOICE ENVIRONMENT IN DEVELOPMENT
“Why do I feel chained to my desk?”— White-collar America
“If it’s not on my phone, I don’t have a desk in my truck.”— Blue-collar America
Bring your compute anywhere you can bring your voice. Talk or gesture; Voice Environment nodes listen and respond, with or without spending tokens. If yesterday’s compute fit in the palm of your hand, tomorrow’s could be in your headset. Connect with your people, direct your agents, control your devices, and choose your environment.
The idea is to blend modern tools into the world around you instead of staying jacked into the Matrix—or, more realistically, glued to a keyboard and monitor. Leave behind the backaches and wrist pain. Stretch out, walk somewhere quiet, or go anywhere you choose.
In development, follow our contact page for potential early access.
Why does my car squeal like a pig until it warms up?
Chat shim — 20B OSS Local RAGCould be belt slip in the serpentine drive; a weak tensioner or misaligned pulley can also squeal.
From Browser, Mobile, endpoint device, or phone number.*
*Some of these are simply proposed client modalities and may not be supported in ultimate releases, pending testing. Phone number endpoints and some mobile OSes may be supported sooner rather than later, or may not be supported, depending on changing or existing app store terms.
VUI is a hub between speech-to-text, text-to-speech, and utilities and applications like messengers that can be located remotely or co-located locally on the client device. It creates deterministic pathways for natural language control.
Natural voice responses are generated with Kokoro. Use command mode to ask for summaries, reread sections, or explanations with the AI/LLM of your choice. Push-to-talk conversations appear in text and voice, which helps when you’re unsure about the transcription or intent.
EVERYDAY USE / TECHNICAL VIEW
Explicit controls, Expected outputs.
Open or set up the client as a background application, then press Speak, Play/Pause, or your chosen hotkey, vocalize your prompt, and press again to submit.
In a typical setup, the connected agent device reaches out to a speech server to convert your voice to text. The text is submitted to the agent, and a probable stream of next tokens ensues. The agent probabilistically determines that you are mostly using it as a fuzzy find-and-replace that corrects spelling and grammar, and maybe even changes your intent, depending on the temperature. A summary of the output is sent to a service that turns text into speech.
Spoken and text responses are sent to you. Command mode lets you change models and speed, repeat a response, explore unsummarized details, and adjust other settings through natural language queries.