Multi-agent systems — a reading list
The papers, framework docs, and posts behind the nine architectures in this project. A starting point for going deeper.
Start here
The clearest overviews of how to think about agents and multi-agent patterns.
The canonical taxonomy — prompt chaining, routing, parallelization, orchestrator-workers, evaluator-optimizer. The backbone of this project.
Single-agent vs. multi-agent, when to add orchestration, guardrails.
A production lead/subagent (hierarchical) system — token economics, orchestration, evaluation.
Foundational papers
The research the patterns are built on.
The reason-act-observe loop underneath tool-using agents.
Self-critique and revision — the idea behind the evaluator-optimizer (v4).
Sample many reasoning paths, take the consensus — the basis of v8.
Multiple agents debate to better answers — the basis of v5.
The original argument for adversarial debate between agents.
Content-driven control over a shared workspace — the basis of v6.
Market/auction task allocation from multi-robot systems — the basis of v7.
Frameworks & SDKs
Tools for building agentic systems.
What this project is built on — providers, tools, streaming, the Experimental Agent API.
Graph-based agent orchestration with durable state.
Multi-agent conversation framework — the choreographed/peer model.
Role-based agent “crews” with task decomposition.
Lightweight handoff-based multi-agent orchestration — the basis of v9.
A WebSocket-bus harness where each responsibility is a swappable worker (see issue #10).
Patterns & posts
Practical write-ups on coordination patterns.
The biology behind swarm coordination (v9).
By pattern
The references attached to each architecture in the app.