Cette page n'est pas encore traduite. Vous lisez l'original en anglais.
Start here
You want an agentic team. Many people talk about doing this. We do it, and this page says where to start. You do not need the big version first.
Jump to
If you just want a website, go straight to the buying steps.
Begin with a very simple workflow and grow only when it earns it. Everything we write here is open source, with no lock-in.
Already run agents and need to control them? Go to riskmandate.ai, which writes the behaviour policy and sells on the same ladder. Come back here when you want a team set up with its policies.
I want a website: the first workflow
This is the shortest route from "I want one" to a person reading your request. It is a proposal of the steps; the end-to-end flow is being built now, the first orders will be run by hand, and nothing has been sold yet. The price is £500, one-off, first proposal, not an offer; Dinis has not yet confirmed a launch price.
- Look at the examples. The hub has a page for each site we have made, all demos.
- Gather what we ask for. A short description of you or your project, in your own words; photos; and videos if you have them. Some of it can be public already, such as an existing profile; send the link. Nothing else is needed to start.
- Email it. Open an email to agent@riskmandate.ai with the subject and three prompts already filled in. Attach files or send links. There is no form: the site stores nothing of yours, and neither do we keep your files after handover.
- A person replies. The email goes to a person, who reads it and answers. We have not set a reply time, so none is promised. The reply says whether your material is enough, which price applies, and what happens next.
- We build it. A team of agents builds the site from your material and shows you a preview. You say what to change.
- It is handed over. The site is a static site in a GitHub repository that you own, and you can change it on your own Claude or host it anywhere that serves a GitHub repository. Our part ends there: no subscription and no support from us.
Want a second opinion before you write? The trust and security page says what our agents can reach and how to stop them. More on the offer is on the website cell page.
Step one: one very simple workflow
Pick one piece of work that repeats. Give it to one Claude agent on a loop, with a written instruction, a place to put its results and a person who checks them. That is the smallest cell. It is a proposal for where to begin, not a product.
If you would rather not set it up, the website cell is the simplest thing we run for you: £250 (launch price, usually £500), one-off. There are no meetings and little of your time. A team of agents builds a public website that you steer from your browser. Nothing has been sold yet, and the end-to-end flow is being built now.
Three very simple workflows to pick from. Each is one agent, one instruction, one place for results and one person who checks:
- A weekly digest. The agent reads the sources you name and writes one page of what changed, for you to read and send yourself.
- A draft inbox. The agent drafts replies to the same few kinds of message. A person edits and sends every one; the agent never sends. Our email cell is an example, listed on the hub as a demo.
- A website for you. The website cell above, and the job-seeker cell, a proposal that does not exist yet.
These are suggestions for where to begin, not products we have sold.
Whatever you run, an agent that sleeps and wakes needs five things: an inbox, timers, memory, permissions and an outbox. We run all five in our own team, as files in one repository.
Step two: add a meeting or two
When one workflow works, a few sessions add more roles, a shared board and a handover into your own accounts. We would prepare, meet you twice, do the work, present it and hand it over. From £1,500, first proposal (prices: The intersection, 2026-11-16).
Step three: a team
Two days of engagement, with more of our time and maybe some travel: a conductor, several agents, a key for each and a board as files. £5,000, first proposal.
Step four: a network
A customer engagement, where we map your work and deliver a whole set of things: several teams of agents, each with its own lead agent, plus policies and a record you can read. £20,000, first proposal.
The cells page has the same four steps, and what each site sells at each step. riskmandate.ai sells agent behaviour policies on the same ladder. If your first need is to control agents you already run, start there.
Ready to talk? The contact is at the foot of this page.
What it looks like at scale
Our own team ran many agents at Web Summit in Lisbon. The counts below come from one public article, Ten days at Web Summit, which counted them from git history and the session list. It says the counts depend on how you count, and that an "agent" means a distinct name in commits.
- Around 250 named agents, across about 20,000 commits in 15 repositories (the article).
- 224 Claude Code sessions, more than 60 of them at once at the peak (the article).
This is our own team's work, not a customer's, and it is a measure of activity, not of quality.
See it working
- This team, run as a cell. How our own team works: its briefs, queues, keys and record, with what failed and when.
- The case study. Read the case study for what was built and measured.
- Examples you can open. The hub lists every cell we have made, one page each, such as the site factory and an email cell. The cells we host are demos.
- Do it yourself. Fork a cell on GitHub is the free way in. Other ways are planned, and that section says which exist today.
Is it safe?
The trust and security page says what our agents can reach, where your data goes, how releases are checked, how keys work and how to stop everything. It also lists what we cannot yet show.
Talk to us
Write to agent@riskmandate.ai. A person reads and answers it. All prices on this page are a first proposal, and nothing has been sold.
See also where we fit in the stack.