Your people are already building software. Give it somewhere to land.
Agents made building cheap. Shipping stayed expensive, so the work piles up on laptops. agenthost turns any of it into a real link, private to whoever should see it, in about a minute.
The bottleneck moved. Almost nobody moved with it.
You gave your teams coding agents, and the cost of building a small internal tool fell through the floor. An analyst can now describe a dashboard and have it working before lunch.
The cost of putting it online did not move at all. That still means a repository, a pipeline, a platform ticket and a wait, which for a two-hour tool is an absurd amount of ceremony. So the tool stays on a laptop, gets emailed round as screenshots, or quietly appears on somebody’s personal hosting account.
The result is an organization producing ten times more working software than last year and shipping about the same amount. Everything in that gap was already paid for. It just never reached anyone.
Days, not sprints
An idea that can be tried on Tuesday gets tried. One that has to be scheduled gets argued about instead, and usually loses to something louder.
The person with the problem builds the fix
The people who feel a bad process every day are rarely the ones who get to change it. Now they can, and they can hand it to their team the same afternoon.
Your platform team gets its queue back
The long tail of small requests stops arriving. What is left is the work that actually needed them.
What it looks like when shipping is free.
Not hypothetical. These are the four things that show up first in every organization that starts doing this.
Internal tools stop being tickets
The rota, the quoting calculator, the dashboard someone rebuilt because the real one takes nine clicks. Built by the team that needs it, live behind their own addresses, changed by asking. How that works →
Prototypes become decisions
A steering group that can open the thing on their own phone gives you a real answer. One watching you drive a demo gives you a polite one. Send a link, not a screen share →
Campaigns ship the week they are decided
The launch one-pager, the event page, the hiring push. On your own domain, with HTTPS handled, without booking anybody’s sprint capacity to do it.
Pilots prove themselves before they cost a quarter
Run the idea in front of real users for a month, then decide whether it deserves the proper build. Your files and your data are yours to take with you when it does.
It runs on the agents you already pay for.
Nothing to install, nothing to standardise on, and no second vendor metering the same work. The agent seats you bought last year gain the ability to deploy.
No token charges from us
Your teams’ agents run on the subscriptions you already hold, so the model spend stays on the bill it is already on. We add no per-token cost, no AI credits and no usage meter of our own. You are not paying twice for the same sentence.
Priced by what exists, not by how hard anyone works
Four euros a month per running app and nothing for people. A team that deploys forty times on Thursday and a team that deploys once cost exactly the same, so nobody is rationing a budget to try an idea.
Any client, no standardisation
Claude, Copilot, Cursor, Codex, Gemini, VS Code. Marketing can be on one and engineering on another, both deploying into the same organization, because agenthost is an MCP server rather than a plugin for one product. Every client that works →
Nothing new to train anyone on
No dashboard, no build configuration, no settings page, and no rollout. Your people already have the interface open, and your agent can also read a broken app’s logs and explain what happened in plain language.
Built to hold a whole organization's worth of small things.
The features that stop this becoming a mess once it is more than one person and one app.
People are free
Unlimited colleagues on the account at no per-seat cost. You pay four euros a month for an app that runs code, with its database included, and nothing at all for the hundredth person who can see it.
Private is the easy option
Lock a tool to the addresses you name, or put one password in front of it. The check happens before your app is reached, so nobody has to build a login and there is nothing for a crawler to find.
A project per team
Group work the way the org chart already does. Unlimited projects, so a department, a client or a campaign can each have their own without anyone rationing them.
Every deploy is reversible
Each version is kept, so a bad change is undone by asking for the previous one back. Nothing anyone deploys is one-way, which is what makes it safe to let them.
Databases come with the app
A tool that has to remember things gets somewhere to remember them, already connected. No second product to buy, no connection string for anyone to handle or paste.
One bill, and an inventory
Everything sits in one organization, on one invoice, and your agent can list what is running on it. The alternative is not zero spend, it is a dozen personal accounts nobody can see.
What legal and IT will ask, already written down.
Published, versioned and identical for every customer, so nothing here waits on us to send it to you.
A DPA that is already in force
GDPR Article 28, live from your first minute on every plan. Nothing to sign, nothing to countersign, nothing to chase through a mailbox.
Read it →European all the way down
French servers, a French operator, a Swedish supplier, no region to configure wrongly, and the one transfer out stated in full.
Where it runs →How it is actually built
Isolation, credentials, transport, deletion. Including the part where secrets never pass through the agent, because a key in a transcript is a key you published.
The security page →Where it isn't the fit.
Short and unsoftened, because finding this out in your own security review is worse than reading it here.
- No SLA, no ISO 27001, no SOC 2. If uptime is contractual or a certificate is a gate, this is a no rather than a conversation.
- No SSO or directory sync. People sign in with a one-time code to their address, so joiners and leavers are not automatic.
- Roles are barely a boundary today. Only inviting people is role-gated, so treat an invitation as full trust and use app-level invitations for anyone who should merely look.
- No private link to your internal network. An app here reaches your own systems only if they are reachable from the internet.
None of which stops it being the right place for the hundred small things. It does mean your core banking system stays where it is, which was never the offer.
What gets asked internally.
Isn't this just shadow IT with a logo?
It is the opposite trade. The tools are being built either way. This is the version where they sit in one organization, on one invoice, in a jurisdiction you have already cleared, instead of on personal accounts you will never find.
What do we do about the ones that succeed?
Move them, and treat that as the good outcome. Your files and your database are yours to export at any time, in nothing proprietary. The pilot having been cheap is the point.
Who supports the thing an analyst built?
The agent that built it, which can read its logs and explain the failure in plain language. That is a genuinely different answer from the one a no-code tool gives you, and it is why this scales past the first enthusiast.
How do we start without a procurement cycle?
Ten static sites are free forever with no card, and a paid account is twelve euros a month. Put one real internal tool on it, watch what happens to the queue behind it, and decide from there.
The rest of it.
Is there an enterprise plan?
No, and we would rather say so than invent one. There are two plans, both published with their prices, and the paid one already includes unlimited people, unlimited projects and unlimited domains. What an enterprise tier normally adds is SSO, an SLA and a negotiated contract, and we have none of the three.
Can we negotiate the contract or add our security annex?
No. One agreement, published, the same for everyone. A term we could not really keep would be worth nothing to you, and section 4 of the DPA says that in the document itself rather than only on a marketing page. You do have audit rights on the terms set out there, and we answer specific written questions within thirty days.
How do we know what our people have deployed?
It all sits in one organization, so ask your agent to list the projects and apps on it and you have your inventory. Behind that there is an audit log: every change, every sign-in and every refusal is recorded with who took it, which agent they used, where they were, and how it ended. What is not there yet is a way for you to read it yourself. Ask us for your account’s record and we will send it, which is a worse answer than a page you can open and a better one than the log not existing.
What happens when someone leaves?
Their work stays, because it belongs to the organization rather than to them. Revoke what they hold immediately yourself: API tokens by id, their agent’s authorisation, their access to any app they were invited to. Then email us to take them off the organization, which is the one step that is not a tool yet.
Do we have to pick one agent for the whole company?
No, and that is a real advantage of connecting at this layer. Marketing can be on one client and engineering on another, and both deploy to the same organization. agenthost is an MCP server, so it works with whatever a team already has.
Who do we talk to?
support@agenthost.eu, which reaches the people who build agenthost. No sales team to be routed through and no qualification call. Send the questions that actually matter and you get written answers.
Start with the thing that has been waiting longest.
One internal tool, one prototype, one microsite. Ten static sites are free with no card, and an app with its database is four euros a month.