Local by design.
A local runtime on Apple Silicon, with speech processed on your Mac.
Chat, speak, research, create, and work with your computer.
One local AI environment. A whole new way to move.
A thinking partner, a creative workspace, and a bridge to your tools. KIRA brings the pieces of your work together.
Explore ideas in text or talk with KIRA Live 1. Keep a conversation going, change direction, and make space for the next thought.
Talk through your idea, ask a follow-up, or switch to text. Your conversation stays connected in KIRA.
A local runtime on Apple Silicon, with speech processed on your Mac.
KIRA checks permissions before actions that change files or your computer.
Documents, slides, research, and code become tangible parts of your work.
KIRA Live 1 is KIRA’s model for everyday chat and native live voice. Speak naturally, interrupt a reply, and continue the conversation without starting over.
When an idea becomes a task, KIRA connects the conversation to research, creation, and computer workflows through its permission-aware runtime. Orchestrator V1 remains available as a separate model for computer work.
Explore the live experienceKIRA listens during playback. Your speech can stop its reply so a new thought or correction can take the lead.
Local conversation history and retrieved context help carry earlier turns into the next step of your work.
KIRA’s guarded tool runtime supplies pathways for research, artifacts, projects, and apps—and checks evidence before claiming completion.
A conversation. A plan. A next step.
Play an example to explore the flow.
Illustrative preview. No microphone access or real computer actions.
Speech runs locally on your Mac. Raw microphone audio is not stored by default.
Explore Live 1’s architectureKIRA brings its models and guarded tools into one environment. Choose the experience that fits your task.
Everyday text chat and native voice. A local Listener, Thinker, and Talker support conversation, speech, and interruption.
Live uses coordinated Listener, Thinker, and Talker weights. The Thinker adds routed residual experts for conversation, tools, memory, workflow state, and permissions. This is a coordinated system, not a single jointly trained tensor.
A separate planning and tool-routing model for KIRA’s computer workflows, paired with runtime permissions and result verification.
Explore the modelThe next Orchestrator chapter is under development. Training is currently paused while KIRA Live 1 is developed.
Follow the buildBuilt for Apple Silicon.
Open source. Ready to explore locally.
Follow the people and possibilities behind KIRA.
Live 1, Orchestrator, and everything that comes next.