Sub-agents

Sub-agents are the specialists in your flow. Each one handles a focused domain — a triage router, a billing expert, a knowledge-base researcher. Together, wired through connections, they form the multi-agent system that handles a conversation from start to finish.

Adding a sub-agent

Click + Add Sub-Agent in the toolbar below the canvas header. A new node appears on the canvas. Select it — the right panel switches to the sub-agent editor where you configure everything about it.

The sub-agent editor

Name and routing alias

Name is the display label shown on the node card and throughout the UI.

Routing alias is the stable identifier the model uses when it decides to transfer or delegate to this sub-agent. It's auto-generated from the name as a lowercase slug (for example, billing-specialist) and cannot be changed once set. Choose a name that clearly describes the sub-agent's role — the routing alias is what the orchestrator sees when deciding who to hand work to.

Description

A short description of what this sub-agent does. This is surfaced to the model as context for routing decisions, so write it as a clear statement of the sub-agent's responsibility rather than a technical note.

"Handles billing questions, subscription changes, and invoice requests."

System prompt

The instructions that shape this sub-agent's behavior. Write it for the specific role: what the sub-agent knows, how it should respond, what it should and shouldn't do. Sub-agents with narrow, well-scoped system prompts route and respond more reliably than those with broad, catch-all instructions.

Model override

By default, sub-agents inherit the model set on the root agent. If a particular sub-agent needs a different model — a faster one for lightweight triage, or a more capable one for complex reasoning — set an override here using the provider and model dropdowns.

Leave both dropdowns blank to inherit from the root.

Execution target

Sub-agents are local by default — they run in the same flow as the root agent. Two other execution targets are available:

Internal A2A — routes to another flow in your organization as if it were a sub-agent. Select the linked agent from the dropdown. Anter handles the handoff in-process; the sub-agent appears local to the orchestrator but its full graph runs independently.

Remote A2A — routes to an agent running on a different platform or organization via the Agent-to-Agent protocol. Provide the remote agent's base URL, project ID, and API key. Use this when you need to delegate work to an external AI service.

Skills and tools

Assign skills and tools to a sub-agent to extend what it can do:

• Skills — domain instruction bundles that load on demand. Assign skills from the catalog using the skill selector. The sub-agent decides which skill to activate based on the conversation topic.
• MCP toolsets — tool catalogs from connected MCP servers. Select which toolsets this sub-agent has access to.

Skills and tools configured here are available only to this sub-agent, not to others in the flow.

Setting the entry sub-agent

One sub-agent in every flow is designated as the entry point — it receives every incoming message first. It carries the ENTRY badge on the canvas.

Check Set as entry sub-agent in the editor to designate the selected sub-agent as the entry point. Changing this setting moves the ENTRY badge to the new sub-agent immediately.

The entry sub-agent is typically the one responsible for triage or initial routing — it reads the incoming message and decides where to send it.

Deleting a sub-agent

Click Delete Sub-Agent at the bottom of the editor. Anter asks for confirmation before removing the node and all its connections from the canvas. If the deleted sub-agent was referenced in another sub-agent's transfer or delegation list, those references are removed automatically.