Your agents, to each other
Agents that hand work to the agent that knows it
Your agents, exposed to each other as tools. A generalist can hand a billing question to the agent that actually knows billing, instead of being given every billing tool itself.
What you get
Agent delegation capabilities, under your rules
- ✓ One tool per deployed agent, carrying that agent’s purpose
- ✓ An agent can call exactly the agents whose tools you enabled on it — nobody by default
- ✓ A delegated run is a complete run: its own record, policy checks and approvals
- ✓ The called agent runs with its own tools, never the caller’s
- ✓ An agent already working further up the chain cannot be called again
- ✓ A depth limit — two by default, up to five
What you need to connect it
- One Agent delegation connection for the workspace — nothing to authenticate
- Deployed agents, each with a real purpose written on it
Discovered tools
What a Agent delegation connection turns into
discovery · Agent delegation
ready- ask_billing_agent execute
- ask_support_agent execute
Discovery names each tool ask_ and the agent’s name. A delegation tool is an action, so a person needs execute permission to trigger one at all.
Human approval
Destructive work waits for a person
approval
waiting- Tool
- delete_customers
- Operation
- DELETE
- Runs against
- Production PostgreSQL
- Agent
- Billing Agent (asked by Support Agent)
Exactly these values
{
"filters": {
"id": 100
}
}
Your decision applies to these values only. If anything about the call changes, it stops for approval again.
If the called agent hits something needing approval it pauses, and the approval appears in the queue against the delegated run.
Connect Agent delegation to your first agent
Connect one system and read the agent Caterfli writes from it. There is a demo connection too, if you would rather look before you wire anything up.