Agent Base
Where a specialist's agent becomes something anyone can run in one click.
Agents have no distribution.
Every good agent is on one person's laptop. Someone spent three weeks getting it right, and there is no way to hand it to the next person.
What gets shared is the prompt — the recipe, not the kitchen. So the next person starts over, the specialist answers the same question again, and nobody's work compounds.
An agent doesn't fit in a text box.
Ask someone for their agent and all they can send you is the prompt. The prompt is the smallest part of it.
The tools
Installed, versioned, working together — not a list of names.
The files
The examples and data it learned the job from.
The browser
Logged in, with real sessions — not a screenshot of one.
The setup
Set up the way that finally worked, after a hundred tries.
So the prompt is what gets sent, and the next person rebuilds everything else from scratch.
Agent Base sends the whole machine, not the prompt.
The agent gets its own computer on hardware you own. Everything it needs lives inside — so the whole thing becomes one file you can hand to somebody else.
Create
Build the agent on a real computer. Its tools, files, browser and setup, all in one image.
Run
It boots on hardware you own and answers on your own network — a browser, a phone, a screen on a desk.
Sell
That same disk becomes a listing. Signed, versioned, priced, installable in one click.
Create and run is the product. Distribution is the business.
A marketplace where the listing is the machine.
Build
Get one agent right, once, the way you already do.
Publish
Its whole disk becomes a listing. Signed, versioned, priced.
Install
One click and it boots on the buyer's machine, already doing the job.
3 weeks → 1 click. The specialist builds it once. Everybody else installs it. That is where the work finally compounds — and we take 20% of every sale on the shelf.
How it runs on your own computer.
A launcher
A small program on your computer. It starts and stops agent computers.
A whole virtual computer
Sealed off, with the agent built into it. Its programs, its browser, its files.
You open it in a tab
From a laptop or a phone. What you are looking at is the agent's own computer.
Why now: every computer people already own can run a second computer inside itself, for free — Mac, Windows and Linux all ship it. And the models only recently got good enough to do the job instead of describing it. Two years ago there was no agent worth selling, and no cheap way to deliver one.
Anybody who is good at something. Agencies go first.
What makes an agent good is knowing the trade, not knowing the model. We cannot hire that. We have to reach the people who already have it — and they publish because they earn, not because we pay them.
They have the trade
Knowing the job is theirs, not ours.
They already sell
No market to teach. A new shelf for what they already price.
Building twice is their problem
Not a favour. Publishing once beats rebuilding per client.
Each arrives with clients
One publisher lands many installers. That is the whole cold start.
This is what the 35% marketing line buys: getting in front of these people where they already work, so the shelf is something they have heard of before they need it.
The machine is built. The shelf is opening.
- Signed, notarized desktop app on macOS and Windows
- 10 releases published through the pipeline
- Export, sign and upload a whole agent disk, end to end
- Stripe live: a purchase grants, a refund revokes, a replay moves nothing
3 users — private beta, invite-only, running agents on their own hardware today.
Agents on the shelf, installs and publishers are the three numbers this round is meant to move, and the only three worth reporting next time.
We sell the machine. We tax the shelf.
$0
One computer.
$50 / mo
Unlimited computers. $500/yr annual.
$250
Paid once. A launch offer, for a limited time — priced to bring the first users in, and closed once the shelf is stocked.
20% of every agent sold on the shelf — the line that grows without us building a single agent.
The arithmetic, not a slice of a pie.
What this round buys
100 agents published
25 publishers earning
1,000 installs
by February 2028
Marketplace revenue at that point is small on purpose: this round buys proof that the shelf moves, not a revenue line.
Three years out
1,000 agents × 200 installs × $99 × 20%
= $4.0M marketplace
10,000 subscribers × $50 / mo
= $6.0M subscription
≈ $10M ARR
The lever is installs per agent, not the number of agents. And once an agent is a server on hardware you own, anything can be its client — a browser, a phone, a ten-dollar screen on a desk. The shelf sells the agent; the clients multiply on their own.
Why us, for this.
Hector Ramos
Founder — BSc in Computer Science, and three years building AI solutions for businesses across multiple industries.
Platform engineer
New hire. Install and publish have to be boring before anybody trusts the shelf.
Publisher engineer
New hire. Sitting with the first publishers while they port the agents they already have.
Built alone: the app on two platforms, the agent image pipeline, the publish path, and payments that survive a refund and a duplicate charge notice.
What we need
$1.2MSAFE, $10M post cap
18 months to fill the shelf: 100 agents published, 25 publishers earning, 1,000 installs by February 2028.
45%
Engineering — two hires. Publish and install have to be boring.
35%
Marketing — reaching the specialists who already have agents, and the buyers who install from them.
20%
Operations — infrastructure, signing, legal.
The risk here is not building it. It is that nobody publishes — which is why marketing is a line and not a footnote.