One system, to run them all
The Agent, the Chatbot, the Operatives and every bespoke build run on one system.
Five parts, and one loop that keeps learning.
Different products, one memory
Each product is a way into the same platform. Teach it through one and the others already know.
Everything it learns becomes connected knowledge
Documents, conversations and outcomes file into one grid. The agent retrieves what is relevant, when it needs it.
Every Opafex agent starts here.
One intelligence, shaped to your business.
What it is made of
Five systems, fully interconnected.
Each one makes the others smarter.
A continuous execution loop.
Plan and approve, execute and learn.
Intelagence
Your agent learns by doing.
Every interaction builds structural understanding of your business.
You upload documents and connect your CRM.
Your agent knows what you gave it and nothing more.
Autonomy level
The same system, the same five parts.
You choose how much control to keep.
Risk and safety class gate every action. Low-risk work runs; sensitive actions wait for you.
- Customer conversations handled
- Workflows automated
- Insights surfaced
- Integrations connected
- Patterns learned
Your agent builds each of these.
It delivers results.
The Agent is live in beta from £500 a month, usage included.
Get started- OAuth providers
- 59
- Operations
- 512
- Universal connectors
- 4
- Knowledge domains
- 9
What it can do
Ask in plain English. Every product draws on the same set, so a build has them the day it lands.
Dashboards, designed on request
Say what to track and it builds the page while you watch.
Widgets read real data, or a table you fill in.
Databases you can fill in
Durable tables for a CRM or anything else.
Fill them by CSV and turn them into widgets and reports.
Field apps for your team
Mobile apps that work with no signal.
Photos, GPS and forms sync back to your dashboards live.
Schedules and triggers
Say “every Monday at 9” and it schedules the work without cron.
Event triggers fire when a row lands or a task finishes.
Documents in your house style
It learns your format from examples you send.
Any document comes back rebuilt in it, as DOCX and PDF.
Ask for it instead of clicking
Nearly everything on the dashboard, you can ask for in chat.
Same permissions, same approval step.
Who sees what
Split your staff by desk.
Each sees only its own records, in chat too.
It reads the web
It fetches a page and reads it.
A real browser starts only when the page needs one.
Why the work lands
The engine does this, so nothing has to opt in. Made-up claims, dead processes and expired tokens are all handled.
Grounded by architecture
It keeps its working separate from its answer.
It checks every claim and reads back every write.
Durable execution
If a process dies, the task picks up at the exact step.
Nothing runs twice.
Self-healing connections
It watches connection health and ignores one-off blips.
Only repeated failures drop a link.
Cost-aware by step
Simple steps go to a cheap fast model.
Hard ones get a stronger one.
How the pieces connect
One agent is the unit of work here.
It is not the limit.
Your field team captures on their phones and the rows land in a table. A dashboard reads it and an agent acts on it.
One agent can hand work to another and use it as a tool. That stays inside your account, and the platform records every hand-off.
A second agent is another pair of hands. Everything here shares your connections, your approval queue and the record of who allowed what.
All of this runs today. What comes next can wait until it ships.
The work it does
Three things most AI tools can’t do, each one useful on day one.
Never an empty dashboard
Ask for a CRM and it builds the database and dashboard.
It fills them in as you work.
A field team you build by talking
Your team captures forms and photos on their phones, even offline.
It all flows back to your dashboard live.
It works while you watch, and asks first
Step-by-step execution you can follow, pausing for your approval on anything risky.
It verifies every write and never makes things up.
Why you can trust it unattended
Four mechanisms, not four promises
We mean unattended literally. The code enforces each check and you can watch it run.
The platform blocks made-up answers.
After every write it reads the record back.
If it invents an ID, the step fails and runs again.
Built like a payments system, not a demo
Nothing runs twice, because the database decides what happened.
A crash picks up at the exact step.
Nothing risky without a yes
One approval queue, and a floor no setting can cross.
It can suggest more freedom, but you flip the switch.
It learns from outcomes
It remembers what worked for you.
It never trains on your data, and you can read its notes.
Where your data lives
The same answer whichever product you buy. It applies to a chatbot on a public site and to a bespoke build holding your most sensitive work.
It stays in your region
Pin an account to the UK or EU.
Every model call goes there, and it fails before leaving the region.
We never train on your data
It improves by remembering what worked for you.
Your material is nobody’s training set, and you can read its notebook.
We see shape, not contents
A support session gets your structure, not your rows.
Reading records needs a reason and an audit row in your tenant.
Or keep it in-house entirely
Private Mode runs storage, keys and models on your own infrastructure.
FAQ
Related
Opafex use cases
How founders, agencies, shops and field teams put the platform to work.
Opafex integrations
Salesforce, HubSpot and Xero. Plus REST and MCP bridges for the rest.
Open-source tools for developers
MIT-licensed MCP servers, with the source on GitHub.
What is an AI agent operating system?
Single chatbots forget, cannot act and fail on anything risky. An agent operating system adds five layers: identity, knowledge and connections, plus execution and governance.
Make it yours
You don’t need another dashboard.
You need an agent that works as hard as you do.
Create your account and get started in minutes.