Security for Memric pilots
Start with narrow access and reviewable work.
A computer-use agent can interact with real software. Before a pilot begins, we agree on the accounts, applications and data Memric may use, what it may do and where it must stop for a person.
1
A separate computer
Memric works from its own remote Windows computer rather than an employee’s personal machine. The pilot is scoped around the applications and accounts required for one defined task.
2
Human control
Approval points are agreed before the pilot. At those points, Memric stops and waits for a person before continuing. The team can review the result and the record of the run before the scope expands.
3
A record of each run
Memric keeps an action record and screenshots for the run so the team can inspect what happened. The applicable storage location, retention period and deletion process are established before live customer data is used.
4
Questions we answer before a pilot
- 1
Which accounts and applications may Memric use?
- 2
Which data does the task require?
- 3
What actions are allowed or prohibited?
- 4
Where must Memric stop for approval?
- 5
What should be retained so the team can review the run?
- 6
Who should be contacted when the task needs a person?
5
What we do not claim
We do not describe Memric as independently audited, universally compatible or certified against a particular security standard. Deployment, model provider, data handling and support terms must match the actual pilot rather than a generic promise.
Security questions can be sent to jin@memric.ai.
Start with the task and its limits.
Show us the current process. We’ll identify the access, review points and unanswered questions before deciding whether to pilot it.