With AI
Let your AI talk to your agents. A person approves.
Your AI can ask your agents anything. The risky actions are held, and a person of your workspace decides.
Your AI can ask. A person answers.
Bring your own AI: Claude, ChatGPT or Grok talks to the agents you host. A risky action waits for a person of your workspace.
One message to a deployed agent
1. Your AI asks
Remind Mr Lang about invoice 2041.
agents · send2. The agent wants to act
Send an email to lang@example.com: "Invoice 2041 is open since 14 days."
It does not happen yet. The answer is a pending approval.
agents · approvals3. A person decides, in the app
ApproveDecline
The decision is written down. Your AI has no approve action.
And it can read
agents · digestThe last result of a scheduled agent.agents · usageAgents and tool calls a day, against your plan.agents · getState, versions and connectors. Never a secret.
An illustration with a made-up agent. Your AI works with the agents of your own workspace.
What it can do, and what it can not.
It can
- List your agents, with their state.
- Send one message to a deployed agent.
- Read approvals, the last digest and usage.
It can not
- Approve or decline a held action.
- Read a key or a secret.
- Reach agents of another workspace.
Every plan has the door.
Your own AI works with Agent host on every plan, over MCP. Prices per month, VAT included, from the plan table.
- Pro€19
- Premium€39
- Max€79
- Ultra€158
Questions owners ask.
Can my AI make an agent do something risky?
It can ask. The risky action is held and a person of the workspace approves or denies it in the app.
Which model do the agents use?
On every plan your agent runs on your own model key. It is sealed and never shown again.
Which AI works?
Any that can use a remote MCP server: Claude, ChatGPT, Grok or your own agent.
Is a decision recorded?
Yes. The hub records who decided what, and when.