Your work stays local
Our cloud isn't designed to receive your code or your models' responses — your agents connect straight to your providers.
The coordination layer for coding agents
Closed alphaGive the mission. The teams form themselves.
Kansei coordinates Claude, Codex, local models, and terminal sessions across your projects. Agents assemble the right teams on demand, spinning up sessions, delegating tasks, coordinating dependencies, reviewing results, and bringing you in only when your judgment is required.
Kansei is in narrow alpha. No card, no spam — we'll email you when there's room.
You must be 18 or older. Kansei is not currently offered to persons ordinarily resident in the EEA or the UK.
Already an alpha tester? Sign in to download
Runs locally on macOS, Windows, and Linux · bring your own subscriptions and API keys
The bottleneck isn't the model
Your agents got faster. Their handoffs did not.
Running more agents only helps until you become their message bus. Every copied prompt, status check, and stalled question pulls you back into work the agents should be doing together — the walls between providers, projects, and devices are still yours to cross by hand.
Without coordination
More agents. More babysitting.
You shuttle context between windows, ask for status, resolve every question, and notice the quiet stalls. Parallel work becomes a queue with you in the middle.
With Kansei
The team carries the context.
Agents hand off, wait, review, and answer one another through Kansei. You set the mission, inspect any session, and step in only when a decision needs you.
handoffs
Work moves by name, not copy-paste.
zero-request waits
Waiting agents sleep: no tokens spent polling.
durable state
Kansei tracks who is waiting on whom.
control
Open any session. Take command.
Spinning up agents is the easy part — terminals, worktrees, a watch script. Keeping them working as a team takes the hard, invisible engineering: messaging, recovery, visibility, alignment. That layer is Kansei.
Works with what you already run
Put the agents you already pay for to work together.
Claude, Codex, local models, and terminals share one cockpit and one coordination layer. Keep your existing subscriptions and keys — Kansei turns them into a team without marking up your model usage.
Claude · SDK + PTY Codex and ChatGPT · chat + PTY Local LLM · chat Terminal · PTY VS Code and Cursor · editor + MCP
Yours by design
On your own machine. Private. Secure. Yours.
Leverage shouldn't cost you ownership. Kansei runs on your own machine — your code and your models' responses stay local, with your agents connecting straight to your providers. Privacy and security here are structural, not bolted on.
Never proxied
Bring the subscriptions or API keys you already pay for; inference goes straight to your providers. Kansei isn't designed to receive or proxy your prompts or responses.
Per-device keys you can revoke
Remote access runs through an encrypted tunnel into your machine. Each paired device gets its own key, revocable instantly.
Blind by construction
The coordination layer isn't designed to receive your content. It knows who's working, not what they're working on.
Controlled delivery
Authorized sessions communicate freely. During a wait, only expected replies wake the session; everything else queues. Nothing outside your environment can reach in.
On hardware you control
Your machine, your credentials, your control. The leverage is real, and it's entirely yours.
We're building this in narrow alpha.
Join the waitlist and we'll reach out as seats open up.
You must be 18 or older. Kansei is not currently offered to persons ordinarily resident in the EEA or the UK.