One cockpit for every agent you run
All your agents, on one team.
Claude, Codex, and your local models — coordinating across every project, reachable from any device, on your own machine. Single-vendor agent teams stop at one provider, one window, one repo. Kansei is the cockpit that makes all of them one.
Runs locally on macOS, Windows & Linux · bring your own subscriptions and keys
Illustrative demo
A guided recreation of how a mission flows.
Start a session with any agent — Claude, Codex, whatever you run — and tell it what you want. Agents are spun up and down dynamically as the work needs: delegating, peer-reviewing, exploring, conferring back and forth, even across projects. Drop into the World View and watch, or jump into any session yourself.
Illustrative workflow recreation. Real end-to-end demo video deferred until the first signed build.
The shift
Your agents can form teams now. Just not with each other.
The assistants you run have learned to coordinate — but each only with more of its own kind, in one window, on one project. The walls between providers, projects, and devices are still yours to cross by hand. Kansei is the layer that takes them down — turning every agent you already pay for into one team that runs itself.
Single-vendor teams, walled in one window.
Agents coordinate — but only with their own provider, in one terminal, on one repo. The moment work crosses a vendor, a project, or a device, you're the wire splicing them together by hand.
One team across every provider you run.
Claude, Codex, and local agents hand off, review, and answer each other — across projects, reachable from any device. The coordination isn't locked to one vendor or one window. It's yours.
The hard part
Spinning up agents is the easy part.
Opening a few agents is quick — terminals, worktrees, a script to watch for changes. Getting them to truly work as a team is a different problem: messaging, recovery, visibility, alignment — the hard, invisible engineering that decides whether a long run holds together or quietly falls apart. Kansei is the layer that solves it.
Agents that know their assignment, recognize their teammates, and pass work to each other by name.
A waiting agent suspends instead of burning tokens to poll. It wakes only for the reply it's waiting on — other messages queue until it's ready.
A durable ledger of who's waiting on whom, so no request is ever dropped.
Every agent stays open to inspection while it runs — see exactly what each one is doing, and step in to redirect any of them, any time.
Sessions stay warm and recover from interruptions, so a run holds steady for as long as the work takes — not until the first stall.
More than message-passing: Kansei aligns your agents on a shared protocol, so they orchestrate the whole job end-to-end. Prose alone never could.
How they coordinate
Cross-provider hand-offs. No human in the wire.
Give one agent a complex GitHub umbrella — a dozen linked issues — and it can run the whole thing: working out what's safe to parallelize, spinning up an agent per work stream, and driving each to a merged pull request. Or keep it small and let two agents confer on a single change. The shape follows the work, not a fixed pipeline.
They coordinate through DeliveryGate — a waiting agent gates its session on just the replies it needs, and every other message queues until those land, so it's never interrupted mid-thought. It sleeps cost-free until then. And if a hand-off ever stalls, Kansei nudges it back into motion on its own.
Open by design
Watch everything. Touch anything.
You direct a team that mostly runs itself. See every agent's status at a glance — who's working, who's waiting, who needs a decision — and drop into any one to steer it, add a teammate, or shut it down.
The team handles its own back-and-forth — and when an agent has a question, another can answer in your place, so the work doesn't stall waiting on your reply. You set the direction and the guardrails; they only reach out when they need your call. Step in from any device, redirect or hand off, then step back out — the rest keep running. You're in command without being in the loop.
The payoff
One operator. A whole team's output.
When agents coordinate themselves, you stop being the bottleneck. Dozens of workstreams move at once while your attention stays free for the calls that actually need you — the throughput of a team, running on its own.
Many projects at once
Run a team on every project you own, all advancing in parallel — your output is no longer one thing at a time.
Reclaim your attention
No more babysitting a prompt. The team works while you think, review, sleep, or build something else entirely.
Scale past your own hands
Command an entire team's output on your own. The leverage of many, in the hands of one.
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.
Your work stays local
Your code and your models' responses never touch our cloud — your agents connect straight to your providers.
Never proxied
Bring the subscriptions or API keys you already pay for; your inference goes straight to your providers. Kansei directs your models without ever proxying your prompts or responses.
Per-device keys you can revoke
Remote access runs over an encrypted Cloudflare tunnel into your machine — managed entirely by your own system, with zero access on our side. Each paired device gets its own key, revocable instantly.
Blind by construction
The coordination layer can't see your content. It knows who's working, never what they're working on.
Only authorized sessions reach in
A session can be messaged or woken only by others you've authorized. Nothing outside your team can send in or interrupt.
On hardware you control
Your compute, your credentials, your machine. The leverage is real, and it's entirely yours.
Works with what you already run
Your agents. Your models. One team.
You already pay for Claude and Codex — Kansei puts them on one team, driving each through the provider's SDK or as a live interactive terminal. Add local LLMs and plain shell sessions too; whatever you run plugs into the same team.
— session list recreated from the product · real chat-view capture deferred until first signed build
On every screen
Drive it from your editor, your code, or your phone.
Bring the full dashboard right into VSCode or Cursor, or wire Kansei into your editor and agents over MCP. Then pair your phone or tablet with a single QR scan and command the whole fleet from anywhere — the same control, on every screen you own.
Pricing
Run it free. Step back with Pro.
Free, you drive. Pro, it drives itself. Free is a real product, not a teaser — the full World View, mobile, and hands-on control of every agent across every provider. Pro is the leap to autonomy: agents coordinating across your fleet, runs going unattended for hours, the whole thing driven through orchestration instead of clicks.
Free
- Full World View, telemetry & mobile
- Run & watch any agent — Claude, Codex, local & terminal
- Tags, grouping & hands-on control
- Bring your own subscriptions · up to 3 sessions
Account required · no credit card
Pro
AutonomyMonthly-only launch. 14-day free trial, card required; Stripe shows the registered monthly price before auto-renewal.
- Multi-agent coordination — DeliveryGate, peer review & agent spawning
- Delegated questions & cross-session routing
- Long unattended runs — cache warming & endurance
- Automate the fleet over MCP · more sessions & projects at once
Stripe-hosted billing when enabled · single monthly plan at launch
Enterprise
- Everything in Pro, for your whole team
- Security, compliance & deployment tailored to your org
- Public Enterprise CTA and terms are pending owner approval
Questions
Do I need a credit card?
Free requires a Kansei account but no card. Pro checkout is hosted by Stripe; any trial and payment-method requirement are shown from the registered launch configuration.
Why pay on top of my Claude or Codex subscription?
Kansei never touches your model bill — you bring the subscriptions you already pay for. Pro buys the coordination layer: agents that hand off, review, and run unattended across providers and projects, all from one cockpit.
Is Free actually useful?
Yes — the full World View, mobile, and hands-on control of every agent.
Do you charge for model usage?
No. Kansei does not resell model credits or proxy your inference. You bring the subscriptions you already pay for; account-backed features may record content-free usage totals for your Kansei dashboard and service protection.
Can I cancel anytime?
Pro billing actions are handled in the Stripe Customer Portal when billing is enabled. The site names Free or Pro only from the active uplift-id entitlement state.
Which platforms are supported?
The cockpit runs on macOS, Windows, and Linux, and pairs with any modern phone or tablet over an encrypted tunnel into your own machine.
管制 control · 感性 sensibility · 完成 completion
Set the mission. Step back.
Start free with an account: point Kansei at a project and put Claude, Codex, and local agents on one team — on your own machine, local, private, and yours. When you want them to coordinate and run themselves, that's Pro. macOS, Windows, and Linux.
Free and local · bring your own subscriptions · macOS, Windows & Linux