Multi-Agent Orchestration Frameworks

Multi-agent orchestration is the practice of coordinating several AI agents toward a shared goal. An orchestration layer decides which agent runs when, what context it receives, what tools it can use, and how failures are handled. The most common structure is an orchestrator that splits a goal into subtasks and assigns them to specialist sub-agents; other styles route work by graph edges, hand control from one agent to another, or let agents talk in a managed group chat. Frameworks differ mainly in how that control flow is expressed, so this hub groups them by orchestration style rather than ranking them.

Multi-agent designs are not always better. Anthropic reports that its orchestrator-worker research system used far more tokens than a chat and suits tasks that parallelise well, while Cognition has argued that parallel agents writing to the same work product can lose shared context and has described narrower designs that work. Studies such as "Why Do Multi-Agent LLM Systems Fail?" document common failure modes. Vendor articles, including the TrueFoundry overview this page draws on, also point out that frameworks do not by themselves provide access control, cost limits or audit-grade logs; see the gateway, governance and observability pages linked at the end. Listings are grouped by style, not ranked, the list is not complete, and framework status changes quickly, so check each project for its current state.

Frameworks where you define agents and their transitions as a graph or workflow, with explicit control over routing and state.

Frameworks that model agents as team members with roles, goals and delegation.

Frameworks where agents coordinate by exchanging messages, often under a manager.

SDKs where one agent hands control to another, or calls other agents as tools.

Infrastructure for running long-lived or large-scale agent workflows reliably.

Open protocols for agents to reach tools and each other across frameworks.

Primary articles and papers on when and how to build multi-agent systems.

Agent frameworks, patterns, protocols, gateways, governance, observability, memory and security pages that relate to multi-agent systems.