The first run is the rehearsal, on real data.
You type the task the way you would say it to a colleague. fluctor plans the steps, names the cost, and runs them. Every step that would leave the house waits for your yes. Then the steps it took are kept as the agent.
Every Monday, remind customers whose invoices are more than 30 days overdue. Same tone as our last reminders.
Four steps: find invoices over 30 days · check the delivery note is on file · draft the reminder · send it. Expected cost for this run: 4 to 9 credits.
Found 3 invoices over 30 days: Halden Steel Ltd, Marlowe Freight, Corbett Fasteners.
Delivery notes on file: 3 of 3.
Reminder drafted for Halden Steel Ltd, invoice_2041, €12,400, 37 days.
Waiting for your approval before it is sent.
Approved.
Recorded as an agent. The same four steps run next Monday at 07:00, and the Monday after that.
LIVE · AGENT RUNS
/agents
Product screenshot
RUN HISTORY · APPROVALS · COST PER RUN
Every run, every approval and what it cost, in one list. A click opens the steps it took.
Five ways a run can start.
The recorded steps stay the same whichever door it comes in through.
A schedule
Every Monday at seven, the first of the month, or a cron line of your own.
A signal
The moment fluctor flags something: a delivery with no invoice, a deadline coming up.
An email
A matching message lands in a mailbox fluctor watches.
A webhook
Another system calls it the moment something happens there.
One click
You run it yourself, whenever you want.
Everything it can do, in one list.
An agent only ever picks from these 30-odd actions. Nothing hidden behind them, nothing it makes up on the way.
Read the memory
- Look anything up: documents, customers, orders, signals, cases
- Show the evidence behind a fact, with file and page
- Sums, breakdowns and top lists from your data
- Read a file
Reach outside
- Draft and send an email, to recipients you allow
- Call another system, read or write
- Open a website in a real browser, fill in a form, read a page
- Search the web, with sources
- Write a file, or bring a new document into the memory
Raise and record
- Open or update a case
- Raise a signal, with its evidence
- Set a watchdog: when this signal appears, start that agent
- Start another agent as a sub-run
- Build or edit a dashboard
Judge and forecast
- Judge a situation, draft a text
- Filter, sort and group a list
- Forecast a metric, run a what-if, score a risk, backtest
- Wait for a signal, or pause until a set time
Three levels of freedom, and you set the ceiling per agent.
Exactly as recorded
Every run repeats the recorded steps, nothing else. If a step fails, the run stops and tells you. The safest setting for anything that has to be identical every time.
Recover on its own
The same recorded steps, but a step that fails may be retried once with different inputs. Still your steps, with room to get past a hiccup.
From a briefing
No recorded steps, only your instructions. It decides the next move as it goes, from the same fixed list of 30+ actions, and stops for your approval before every step that reaches outside.
What holds it, run after run.
An agent that sends mail and changes records has to be held to the same standard as the person who used to do it. These are the rules it runs under.
A cost cap per agent
Set a budget for each agent. Hit it, and the run stops instead of continuing to spend.
Checked before it reaches outside
Every message, record change or call to another system is compared with your instructions first. Anything that does not fit stops and asks.
A log of every step
Every step, every decision and every approval is written down with a time and a name.
Four items or four hundred
A run over a list does not stop for one failure. The failed ones are listed by name, and the run is marked as completed with errors, never as a clean success.
A fork it has not seen
A situation the first run never met is handled once with your approval, then kept. The recorded path is never rewritten, only extended.
An agent reads from the same memory your people ask, and reacts to the same signals.
Show us something you do regularly.
Walk fluctor through it on the call, on your own documents. You leave with the agent it recorded.
No integration project to start · Your data stays in the EU