security & control
Security and control
Agents only reach what you give them, only act when your team says so, and leave a record of every step. Here’s how.
what agents can do
You decide, account by account.
Only what you tick
Each agent is given specific mailboxes, accounts and folders, and folders can be limited to reading only.
Approvals before anything important
Choose which actions need a yes. The agent shows exactly what it will do, and does exactly that once approved. Looking things up never needs approval.
Rehearsals
Try a new agent or a change on real data first. Nothing is sent, created or changed, by that agent or any agent it hands work to.
People and roles
Admins manage the workspace and members build and run agents. Custom roles can limit what each person sees and changes.
what happened
Every step is written down.
A record of every job
What started it, every step it took, every question it asked and who approved what, when.
A record of every change
Every edit to an agent is a saved version. Changes made by the Manager or from a connected assistant say so.
A spending cap
A monthly limit for the whole organisation. When it’s reached, agents stop.
your data
Where your data goes.
Never used for training
We don’t use your content to train AI models, and we don’t sell or share it for advertising.
Logins kept encrypted
Website logins you save for an agent are encrypted before they’re stored.
Your history, your call
Your organisation’s owner chooses how long job history is kept, and a workspace can be deleted when you’re done with it.
Hosted, or on your own servers
Polytask is hosted in the United States. Larger organisations can run it on their own servers, using AI in their own Microsoft Azure account.
Start with one job.
We’re opening to a small group and setting each business up personally. We’ll build your first agent with you, rehearse it on your real email and documents with nothing sent, and switch it on when you’re happy.
Nothing important goes out without your OK — from the first minute.