Skip to main content
Sub-agents enable multi-agent workflows. An agent can delegate work to other specialized agents.

​
When to Use Sub-Agents

Use sub-agents when:
  • Task has distinct phases needing different expertise
  • Want to parallelize work
  • Need specialized agents for specific tasks
Example:
  • Main agent receives customer inquiry
  • Delegates to Research sub-agent
  • Delegates to Analysis sub-agent
  • Compiles results

​
Creating a Sub-Agent Call

Agents with access to compatible connector can call other agents: 
Call sub-agent "data_analyzer" with prompt:
"Analyze this dataset and provide insights"
Agent automatically:
  1. Creates execution context
  2. Passes data/context
  3. Waits for response
  4. Uses results in own reasoning

​
Configuration

​
Enable Sub-Agent Delegation

  1. Agents > Select main agent
  2. Settings > Capabilities
  3. Attach Agent Caller connector (if available)
  4. Save

​
Call Sub-Agent

In agent system prompt, add instructions: 
When user provides data analysis request:
1. Delegate to "data_analyzer" agent
2. Wait for results  
3. Summarize findings
4. Provide recommendations

​
Sub-Agent Handoff

Smooth handoff between agents: 
"I'm passing you to our specialist agent..."
(calls sub-agent)
"They found: [results]"

​
Best Practices

​
Clear Specialization

Each sub-agent should have clear, specific role.

​
Error Handling

Handle sub-agent failures gracefully: 
If sub-agent fails:
1. Retry once
2. If still fails, inform user
3. Offer alternative approach

​
Data Passing

Only pass necessary context to sub-agents. Too much context = higher costs.

​
Avoid Infinite Loops

Sub-agents shouldn’t call parent agent. Set recursion limit to 3.

​
Architecture Example

Each sub-agent handles specific type of task.

​
Monitoring

Check sub-agent activity:
  1. Agents > Select main agent
  2. Activity tab
  3. See sub-agent calls and responses

​
Costs

For current pricing details, see Pricing. Sub-agent calls are charged same as regular agent execution. Monitor in Account > Usage dashboard.

​
Delegation Details

​
How Delegation Works

When an agent delegates to a sub-agent, the platform uses a synthetic delegate_agent tool. The parent passes a task description and context, the sub-agent executes with its own capabilities, and the result flows back to the parent.

​
Delegation Depth

Sub-agents can delegate to their own sub-agents, creating a chain. The default maximum depth is 3 levels. The platform tracks lineage (which agent called which) for observability and to enforce depth limits.

​
Shared Work Budget

Parent and child agents share the same work budget. A sub-agent’s token usage and tool calls count toward the parent’s budget. This prevents runaway costs from deep delegation chains.

​
Isolated Threads

Each sub-agent gets its own thread for execution. It doesn’t see the parent’s conversation history — only the task and context explicitly passed to it. This keeps sub-agents focused on their specific task.

​
Limitations

  • Sub-agents can’t see parent conversation history
  • Can’t modify sub-agent behavior mid-call
  • Sub-agent response must be usable by parent
  • Maximum delegation depth: 3 levels (default)

​
Next Steps