The AI agent you talk to
Double-tap left ⌥ and just say it. Kocoro runs the same agent hands-free — schedules, Chrome, email, apps — and ships real files.
Voice and Episodic Memory are Mac-only for now.
KOCORO VOICE
Voice isn't dictation. Double-tap left ⌥ from anywhere and start a call — Kocoro listens, runs the real task with your agent, and talks back. Mute the mic and the task keeps running; hang up with Esc when you're done.
Double-tap left ⌥ to start a call
Not a voice assistant bolted on. Your spoken request runs the same agent — web, files, code, schedules, email — and ships the same real deliverables.
Voice uses tokens only during active conversation — when you speak or Kocoro responds. Sit on the line in silence as long as you like; it costs nothing.
Double-tap left ⌥ to talk, again to mute or cut in, Esc to hang up. No typing, no mouse — manage the call with ⌥ and Esc.
Voice lives in the menu-bar island, floating over any app. Start a call without even bringing Kocoro to the front.
EPISODIC MEMORY
Every conversation trains a private memory model — built only for you, stored on your Mac. Talk to Kocoro and it already knows your projects, your people, the decisions you made last week. No shared memory. No cloud RAG. It learns you the way a long-term colleague would.
Continue the Q3 deck.
A personal model, not a shared one. Your memory never mixes with anyone else's.
Stored locally. No cloud RAG, no vector database someone else can read.
Projects, people, tools, decisions — recalled by name, not re-explained.
Double-tap ⌥ and it picks up mid-thought, Friday's context already loaded.
Episodic Memory is opt-in and off by default.
Slack
企业微信
Lark飞书
LINE
Intercom
Telegram
PtengineMeta
ShopifyNotion
Google Ads
GmailGitHubFigmaMost of a day isn't the hard thinking — it's the steps between idea and done. Here is how Kocoro picks them up, in the app.
Describe a recurring job in plain language — a morning digest at 8:00, a report draft every Friday, a price check every 6 hours. Kocoro runs it on schedule, stays quiet when all is well, and pings you when something needs your eyes.
@mention Kocoro from Slack, Lark, or WeCom, and the task runs on your Mac. Results post back to the same conversation — with a link to every file it made.
@Kocoro pull this week's signups and draft the Friday update
Done — draft is below. File saved to your Mac.
Kocoro works directly on your Mac — reading and writing local folders, checking Apple Mail or Gmail and your calendar, driving a real Chrome window: clicking, filling forms, extracting data. Deliverables land where you asked.
Install community Skills or package your own workflows, then ask for the deliverable you actually need — PPT, PDF, XLSX, charts — written straight to your local folder.
KOCORO ROBOT
Coming soonKocoro Robot brings the same agent into the physical world: a friendly desktop robot, powered by a Mac mini running Kocoro, that greets visitors, routes them to the right person, joins meetings, and takes the notes. Same Kocoro behind the scenes — only the way in changes.
Greets your guests, handles the first questions, and notifies the right person — all hands-free.
Joins the room, captures decisions and action items, and shares the notes with everyone.
Kocoro keeps useful context as structured memory and turns repeatable work into reusable Skills. The result is a workspace that gets faster to brief and easier to share.
“Every good workflow can become the starting point for the next one.”
Conversations, files, and configuration are stored on your Mac by default — not on our servers.
Your files stay on your Mac. AI requests run over an encrypted connection, and neither the LLM provider nor Kocoro trains shared models on your data. Optional cloud features stay off until you turn them on.
Set precise allowlists and denylists for tools. When to run automatically, when to ask first — your call.
“Your Mac, your data, your call.”
Start free. Upgrade when you need more.
Yes. The free tier includes 3M tokens per month with no payment required to start. Upgrade to Pro or Max when you need more capacity.
Kocoro uses a monthly subscription with a token quota. Once you exceed your monthly quota, usage is billed at $5 per 1M tokens. Prepaid credits ($10 · $25 · $50 · $100) are also available and carry over month to month.
Storage is local by default — your chats and uploaded files stay on your Mac. AI inference runs in the cloud: message content travels over an encrypted connection to Kocoro's cloud, then to the LLM provider for model processing. Two optional features — Episodic Memory and Session Share — are off by default and only involve the cloud once you enable them.
Kocoro isn't tied to any single AI provider. Your requests run over an encrypted connection through Kocoro's cloud (Shannon Cloud), which routes them to a managed LLM provider. There is no fully-offline local inference mode in the standard desktop app today.
It refers to storage: chats and uploaded files stay on your Mac by default. AI inference runs in the cloud over an encrypted connection; the LLM provider does not use your data to train its models. By default, Kocoro does not train any AI model on your data either. If you enable Episodic Memory, Kocoro uses that data to train a personalized memory model used only for you — it does not feed into any shared model, and the LLM provider never receives it for training. Using external channels (Slack, Lark, WeCom, etc.), session sharing, or cloud orchestration tasks routes the relevant content through Kocoro's cloud when those features are active.
Yes — for Business customers, Kocoro can run fully on-premises with Ollama and the open-source models it serves. Contact us for a plan: contact@kocoro.ai
Runs on your Mac. Free to start. Begin with a real task and keep control of the result.
Voice and Episodic Memory are Mac-only for now.
Install takes about 2 minutes.
Kocoro — from the Japanese kokoro (心), meaning heart.
