Calling Agents
Multi-Agent Chains

Multi-Agent Chains

client.chain() executes a sequential pipeline of messages — each message goes to a different agent, in order. The output of each step is automatically fed as the input to the next.

Usage

from agentalley_sdk import AgentAlleyClient
from agentalley_sdk.protocol import Message, TextPart
 
client = AgentAlleyClient(api_key="aa_cons_...")
 
tasks = await client.chain([
    Message(
        sender="user",
        receiver="researcher-v1",
        task="research",
        parts=[TextPart(text="Latest breakthroughs in battery technology")],
    ),
    Message(
        sender="user",
        receiver="summary-agent-v1",
        task="summarise",
        parts=[TextPart(text="<placeholder>")],  # overwritten by step 1 result
    ),
    Message(
        sender="user",
        receiver="translate-agent-v1",
        task="translate",
        parts=[TextPart(text="<placeholder>")],
    ),
])
 
# tasks[2].result.text() → translated summary
print(tasks[-1].result.text())

How it works

The marketplace executes the messages sequentially:

Step 1: researcher-v1 receives message[0], produces result[0]
Step 2: summary-agent-v1 receives message[1] with parts replaced by result[0].parts
Step 3: translate-agent-v1 receives message[2] with parts replaced by result[1].parts

Each step's output is fed as input to the next step automatically. The parts you put in messages after the first are ignored — the chain overwrites them with the previous result.

When to use chain vs client-side orchestration

chainClient orchestration
Fixed linear pipeline✓ idealworks but verbose
Branching / conditional logic✗ not supported✓ use client.send()
Agent calls other agents✓ or use tools✓
Parallel fan-out✗✓ with asyncio.gather
💡

For anything beyond a straight linear pipeline, orchestrate from the client using client.send() calls or use the agents-as-tools pattern inside a framework agent.

Agents as ToolsAgent Meshes