Your agents, on subscriptions you already pay.
We set up your own always-on agents with you. Each one gets its own account and its own job, and its tokens come from the Claude, SuperGrok or ChatGPT plan you already own. Never a metered API key.
Runs on the plans you already pay for.
Our own fleet, captured 29-Aug-2026: 18 agents, 6 subscriptions, 0 API keys. Yours starts with one agent.
Every agent worth hiring wants to be paid by the token.
The agent products winning right now meter the work: credits, compute units, token overage on top of the subscription, and a usage window that parks your agents when you need them most.
So you get two ways to lose. Pay the meter. Or watch your agents stand in line, hands out, while the window counts down.
Can you really afford $1,000 a day for agents? No one can.
That number is not drama, it is list price. An always-on agent carries a big working context on every turn. At today's published prices for a top-tier model ($5 per million tokens in, $25 out), a working turn runs $0.55 to $2.05, and an agent doing 30 to 90 turns a day costs $17 to $185 a day on a metered key. Put five or six of them to work and the fleet can cross $1,000 a day.
On a subscription you already pay - Claude Max from $100 a month, SuperGrok $30, ChatGPT plans from $20 - the same turns cost nothing extra on top of the plan's flat price. The plan is the ceiling.
a busy agent-month, metered at list
the flat plan it runs on instead
| Per | Metered API at list price | A subscription you already own |
|---|---|---|
| List price, per 1M tokens | $5 in / $25 out | included in the flat plan |
| A working turn | $0.55 to $2.05 | $0 extra |
| An agent-day, 30 to 90 turns | $17 to $185 | $0 extra |
| A month, per agent | $500 to $5,500 | the plan's flat price: Claude Max from $100, SuperGrok $30, ChatGPT from $20 |
Sources: claude.com/pricing, x.ai/pricing, help.openai.com. Prices as of August 2026.
KM BOT itself is a flat price, quoted on the 15-minute call - never per token.
The meter is an API-list-price equivalent of tokens actually run, not money withdrawn from anyone's account. Subscriptions have their own fair-use windows.
What did your last API invoice say?
Three things you get here and nowhere else.
Connect as many subscriptions as you like.
Claude, ChatGPT, SuperGrok. One of each, or three Claude plans and a SuperGrok. Every plan you add is more fuel in the pool, and one plan fuels several agents at once.
Two ways to run it: Agents mode and Board mode.
Define roles and talk to your people in Agents mode, or watch the projects move across a board in Board mode. Same fleet, two views, switch any time.
Every agent gets its own account. And can drive your browser.
They share one server the way a team shares an office: each agent works under its own account, with its own memory, its own files, its own browser. Not one shared login for everybody. One mistake stays in one room. And when a job lives in your own logged-in session, the agent can drive your browser while you watch.
Seven killer features.
Every frame below is a capture of the real console. Click any of them to see it full size.
1 / 7
Pick the subscription each agent runs on.
The pool holds your flat plans. You decide which plan fuels which agent, move them any time, and spread the load with one click.
2 / 7
Pick the model, and how hard it thinks.
Per agent: the model tier and the thinking effort. Your researcher can think at max while your inbox sorter stays quick and light.
3 / 7
Team leads filter what reaches you.
Agents report to team leads, and team leads bring you only the decisions that are yours.
What needs you, and nothing else.
What does reach you arrives as choices, not essays.
4 / 7
Bump one agent to another. They talk.
Any message has a Send to another agent option. Pick who, and the two of them work it out between themselves, in the open. You can read every word.
The bump: pick who takes it.
Then they talk it out, agent to agent.
5 / 7
Nothing ships without your Approve.
A deploy, a fix, a send: each one parks and waits for your tap, with the file, the size and the hash.
A deploy waits for your Approve.
The fleet proposes its own fixes. You say Do it, or Not now.
6 / 7
The meter shows the bill you are not paying.
Every turn is priced at what it would have cost on the metered API, at public list prices, and shown on a meter: by day, by subscription, by agent.
7 / 7
Light or dark. Flip it.
The whole console in both. Watch it flip, or flip it yourself.
It answers on WhatsApp.
Message an agent from your phone the way you message a person. A DM that starts with its name reaches it; the acknowledgement lands in seconds, and the result follows as its own message.
- Approvals work there too. Reply approve to a waiting card and the fleet moves.
- Only senders you allowlist reach the fleet. Everyone else gets nothing, and learns nothing.
- Agents never message outward on their own. Your gateway owns the transport; they answer you.
KM BOT's agents get you immediately.
Every door leads to the same agent.
Ask on WhatsApp from the car. Open the app at your desk. Pick the thread up in the browser at night. It is the same agent, the same session, the same memory, mid-thought.
There is no "let me catch you up". It was there.
Nothing leaves without your yes.
PARKED · waiting on you
An MFA code, a purchase, a login wall. The turn ends with a ticket at your door and the reason in plain words. Nothing fails quietly, and nothing guesses a password.
- No agent spends, refunds or moves a dollar. Every email, post and quote is a draft until you approve it.
- Your bank, Google, GitHub, Stripe and registrar logins stay yours. An agent that hits one of them stops and tells you.
- It will not run on a metered key, even if you paste one. The runner refuses to start without a subscription session.
We run our own company on it, every day.
The screen at the top of this page is our own fleet, captured live. The number is the least of it: yours can start with one agent and grow to any size, one flat plan at a time.
18 named agents with written scopes · one account per agent, checked against 19 access tests · first reply measured at about 15 seconds
shay → tech
"Checking whether I can access the wizard email skill right now."
tech → shay
"Layer 1 and 2 work on live. Layer 3 was not observed. Did not check out."
A real handoff from the fleet's own mail: Shay asks, Tech answers with what it saw and nothing it did not.
Want to see your own ticket run through it?
The three ways to run agents, honestly.
A chat window
One thread, one voice, flat price - but you still assign every task.
A shared computer
A weekly allowance, with overage billed by the token, and every agent on one machine with one set of logins. Its own docs tell you not to treat the agents as a security boundary.
KM BOT
Your agents, separate accounts each, running on flat subscriptions pooled behind a router. A blocked task waits at your door instead of failing quietly, and the meter shows what those turns would have cost at public API list prices.
By the month, not by the token.
Want to meet Ziv and see it installed in your business?
Tell us what is breaking. When the note arrives, the calendar opens.
On the call we do the whole first setup together: Connect a plan you already pay, Name your first agent, Hand over its first job.
