Skip to main content

Overview

The Swarms API provides a diverse range of multi-agent architectures, each designed to solve specific types of problems and workflows. These architectures enable you to orchestrate multiple AI agents in various patterns, from simple sequential workflows to complex hierarchical systems. Understanding the available architectures helps you select the optimal pattern for your use case, whether you need parallel processing, consensus-based decision making, or dynamic task routing. Multi-agent architectures can be combined and nested to create sophisticated systems that leverage the strengths of different patterns. Each architecture type has unique characteristics that make it suitable for specific scenarios, such as linear processing pipelines, collaborative problem-solving, or high-throughput batch operations. The flexibility of these architectures allows you to build custom solutions that match your exact requirements.

Available Architectures

The following table provides a comprehensive list of all available multi-agent architectures in the Swarms API:
Swarm TypeDescriptionDocumentation
AgentRearrangeDynamically reorganize agents to optimize task performanceLearn More
MixtureOfAgentsCombine diverse specialist agents for complex tasksLearn More
SequentialWorkflowExecutes tasks in a strict, predefined orderLearn More
ConcurrentWorkflowRuns independent tasks in parallel for higher throughputLearn More
MultiAgentRouterIntelligent dispatcher that routes tasks based on capabilities/loadLearn More
HierarchicalSwarmMulti-level structures with delegation and escalationLearn More
MajorityVotingConsensus-based decision-making across multiple agentsLearn More
BatchedGridWorkflowExecute multiple tasks across multiple agents in a grid patternLearn More
GraphWorkflowExecute a graph workflow with directed agent nodes and edgesLearn More
AdvancedResearchAdvanced multi-agent research workflows and strategiesLearn More
GroupChatCollaborative problem-solving through conversation-
AutoSwarmBuilderAutomated swarm generation-
MALTMulti-agent learning and training-
CouncilAsAJudgeCouncil-based evaluation system-
InteractiveGroupChatInteractive collaborative sessions-
HeavySwarmHigh-capacity swarm processing-
LLMCouncilLanguage model council for decisions-
DebateWithJudgeStructured debate with judgment-
RoundRobinRound-robin task distribution-
autoAutomatic swarm type selection-

Architecture Categories

The available architectures can be broadly categorized into several groups: Workflow Patterns: SequentialWorkflow, ConcurrentWorkflow, GraphWorkflow, and BatchedGridWorkflow provide different execution patterns for organizing agent tasks. Collaboration Patterns: GroupChat, MixtureOfAgents, InteractiveGroupChat, and MajorityVoting enable agents to work together collaboratively. Routing & Organization: MultiAgentRouter, HierarchicalSwarm, AgentRearrange, and RoundRobin help organize and distribute work among agents. Specialized Systems: AdvancedResearch, HeavySwarm, AutoSwarmBuilder, and MALT provide specialized capabilities for specific use cases. Decision & Evaluation: MajorityVoting, CouncilAsAJudge, DebateWithJudge, and LLMCouncil focus on decision-making and evaluation processes.

Getting Started

To use any of these architectures, make a request to the /v1/swarm/completions endpoint with the swarm_type parameter set to your desired architecture. For detailed information about each architecture, including usage examples and best practices, follow the links in the table above or refer to the Multi-Agent Overview page.