​SDK Deep Dive: Building Multi-Agent Systems with Summoner

​Description

​In our previous session, we introduced the big picture: moving from single agents to coordinated systems.

​This session goes deeper.

​We’ll walk through the Summoner SDK in detail — how it works, how to structure projects, and how to design reliable multi-agent flows without writing brittle orchestration glue code.

​If you're building agents and want to understand what happens beneath the abstraction layer, this is for you.

​What We’ll Cover

​• SDK architecture and core components
• The orchestration loop (events → handlers → state transitions)
• Async send/receive flows
• Hooks, routing, and message filtering
• Relationship-based state management
• Designing predictable multi-agent interactions
• Extending the SDK for custom logic

​We’ll explore a real project built from the SDK template and walk through the operational semantics step by step.

​This is not a high-level overview.
Expect code, structure, and mechanics.

​Who This Is For

​• Developers building with LangChain, CrewAI, MCP, or custom agent stacks
• Engineers deploying agents into enterprise environments
• Builders curious about durable multi-agent coordination
• Anyone who attended the intro and wants to go deeper

​Format

​• 45 min deep technical walkthrough
• 15 min Q&A

​Prerequisites

​Familiarity with Python and basic agent concepts is recommended.
No prior Summoner experience required.

​Why This Matters

​As agent systems grow beyond one-off workflows, coordination becomes infrastructure.

​Understanding the SDK layer is the difference between:

​“My agent works.”

​and

​“My agents work together — predictably.”