Agents you can trust with real work
An agent that acts for your business must be safe to trust. These are the rules every Maahi agent is built on.
How every agent is built
Least-privilege access
Each agent receives only the access its job needs, through accounts we can limit and revoke.
Approval points
Sensitive actions such as payments, deletions and outside messages can require a person to approve them first.
A record of everything
We keep a log of what the agent read and what it did, so anything can be reviewed.
Clear data boundaries
We agree in writing which data the agent can see, where it is processed and how long it is kept.
Tested before launch
Agents pass a written set of real cases before they touch a live system.
People stay in charge
You decide what runs alone and what waits for a person, and you can change it at any time.
Decide what an agent may do alone
We sort every action into three groups: actions the agent may take on its own, actions that need a person to approve, and actions it may never take.
Reading a calendar is low risk. Sending a quote or issuing a refund is not. The rulebook makes that difference explicit and enforces it.
- Allowed: reading, drafting, booking within limits
- Approval needed: quotes, refunds, outside messages, changes to records
- Never: deleting data, changing permissions, moving money
Four layers between an agent and a mistake
Limited access
The agent can only reach what its job requires, through a dedicated account.
Rules and approvals
Sensitive actions wait for a person, with the context needed to decide.
Logging
Every input, decision and action is recorded for review.
Testing
A written test set runs before launch and after every change.
Questions to ask any AI vendor
Use these with us and with anyone else.
- What can the agent access?You should get a specific list, not a general answer.
- Which actions need approval?Ask to see the rulebook.
- Can I see a log of what it did?You should be able to review every action.
- Where is my data processed and kept?Ask for this in writing.
- How is the agent tested?Ask for the test set and results.
- How do I switch it off?You should be able to pause an agent at any time.
Straight answers
Does the agent see all of our data?
No. It sees only what its job requires, and you can review and change that access.
Can we see what the agent did?
Yes. Every action is logged and can be reviewed.
What if the agent is wrong?
Sensitive actions wait for approval, the agent hands over when unsure, and mistakes are used to improve the test set.
Do you train models on our data?
We agree data use in writing before work starts, and we do not use your data for purposes you have not approved.
Can an agent be switched off quickly?
Yes. You can pause an agent at any time, and its access can be revoked from the connected accounts.
Ready to talk it through?
Tell us the task and we will tell you whether an agent fits.
Request a demo Book a call