Your agent, your way, however you work.
It connects your tools and runs multi-step work behind approval gates you control. It builds its own dashboards and tables and learns your business.
What do you get?
One price, usage included. Everything below comes with it and needs no separate implementation phase.
from £500/month
- 59 built-in providers plus MCP, A2A, REST and your own APIs
- Multi-step workflows that run behind approval gates you control
- Dashboards it designs and binds to real data from your tools
- Durable tables you and it can both fill in, edit and upload to
- Scheduling in plain English, and triggers that fire on a change
- Enough usage included for consistent work all month
How much rope you give it
You set one dial. Raising it moves more work past your approval queue, but two things always wait.
Reading and reversible work runs. Anything that changes your data waits.
Runs on its own
Waits for your yes
These always wait. They form the safety floor.
The agent cannot lower this floor and cannot earn its way past it. If the approval check fails, these actions refuse to proceed.
Approve the same action often enough and it stops asking. That trust comes from your decisions.
It does real work
Ask in plain English. The agent builds, fills, schedules and acts across your tools and your team.
- It designs your dashboards
- Say what to track and it builds the page while you watch.
If nothing holds the data, it makes a table. - It builds you a database
- Ask for a CRM and it builds you a table.
Edit it, upload a CSV, turn it into a chart. - It schedules and reacts
- Say “every Monday at 9” and it schedules the work.
Event triggers act the moment something changes. - It runs your field team
- Add operatives at £20 each, or £30 (minimum 3) without an agent.
Mobile apps that work offline and sync back live.
What it will not do
The platform enforces the limits below. The agent's judgement plays no part.
- It cannot quietly make something up
- It keeps its working separate from its answer.
It checks every claim against real data and reads back every write. - It cannot act beyond what you allowed
- One dial: manual, guided or autonomous.
Anything gated waits in one queue for your yes or no. - It does not lose work
- Execution is durable and crash-safe.
An interrupted task resumes at the exact step, with no double-runs. - It does not act on a broken link
- The agent tracks connection health continuously.
Only repeated failures drop a connection and it notices recovery itself. - It does not overspend on easy work
- Simple steps go to a cheap fast model.
Hard ones get a stronger one. - Your data can stay yours
- Private Mode keeps your storage, keys and models on your own servers.
Or pin every model call to the UK or EU.
How it works
One loop of five steps. Your agent gets smarter every cycle.
1
You talk
Chat or dashboard.
Same intelligence, your choice.2
Agent reasons
Your agent maps the request to tools and workflows
3
You decide
Stay hands-on or let it run.
You set the autonomy level.4
It executes
Real tools, real integrations, real results
5
It learns
Outcomes improve future behaviour
The dashboard is the same code as this page
These are the product’s own widgets, not pictures of them. They are built from the same components your agent puts on your dashboard, so what you see here is what it renders.
Volume, and what it escalated
A run, step by step
This morning
Waiting on you
Nothing in this column ran on its own. The agent proposed each one and stopped; a person clicks, and only then does it act.
The components are real and the numbers in them are made up. We are not going to put a customer’s data on a marketing page.
FAQ
Get started
Your agent, working by this afternoon
From £500 a month, usage included.
Create an account and it sets itself up around how you work.