Build your own custom agent
How we do it
Choose the agent
Start from one of our agents, or from the work you want an agent to do. We settle which systems it touches, what it should cost to run and who signs off.
You get
A scope
Which agent, which systems it touches, and what it will cost to run each month
Connect it safely
Access is granted for the work and limited to what the agent needs. Nothing runs until that is written down.
- Read-only first
- A service account, not a person
- Only the data it needs
- Consent, then revoke
- Quebec data handling
- A separate space to test
You get
An access plan
What the agent can reach, under which account, and until when
Test it
Before it goes near live data, we check it against real examples from your work: does it get the right answer, stay inside its access and stop when it should.
- Real examples from your work
- Accuracy measured
- Edge cases and failures
- Access limits checked
You get
A test report
What was tested, how accurate it was, and what failed and was fixed
Run it and hand it over
The agent does the repetitive reading and drafting, and people decide what changes. It is watched after launch, and a custom agent is handed over with the documentation to run it.
- A person approves each change
- Handover
- Monitoring
- Limits and a pause
You get
A monthly review
Cost, failures and accuracy, and who approved each change