AI Agent Tools With Multi-Agent Support
24 of the 94 tools in our index document a real multi-agent capability — but read the sourced facts behind the tag and it hides five structurally different things: composable orchestration frameworks you design yourself, no-code team builders, vendor-built agent teams, a single task fanned out across parallel agent copies, and live voice handoff mid-call.
On this page
A framework wiring multiple agents into a graph, a voice agent transferring a call to a specialist mid-conversation, and a coding agent running a fleet of parallel copies of itself are all tagged `multi-agent` on this site, and none of them do the same job. Our own census, [What 'Multi-Agent' Actually Means](/blog/what-multi-agent-actually-means-2026/), read the sourced facts behind the tagged listings and sorted them into six shapes; this collection is that finding turned into a browsable set. Membership is re-derived from each listing's own text every time this page is re-read, never from the tag alone.
Five shapes have members this round. The sixth — *multi-vendor agent hosting*, where the product's pitch is running *other vendors' competing* coding agents side by side — currently has no Published listing whose own sourced text clears the bar, so the shape is recorded here with zero members rather than dropped from the methodology.
**Composable frameworks (9 of 14 [frameworks](/category/frameworks))** — you design and wire the multi-agent system yourself: LangChain, LangGraph, CrewAI, Microsoft AutoGen, Microsoft Agent Framework, Google ADK, Agno, Claude Agent SDK and LlamaIndex. LlamaIndex is the member a reader is least likely to expect, since it is known as a retrieval framework: it earns its place on AgentWorkflow, which hands control between a declared set of agents automatically, or exposes sub-agents to an orchestrator agent as tools. The frameworks in the index with no shape here are excluded on their own published text: Letta and Mastra name no agent-to-agent coordination primitive, and AgentsKit mentions the A2A spec only as a portability format that keeps agent definitions readable across tools.
**No-code team builders (1 of 16 [agent platforms](/category/agent-platforms))** — a non-technical operator assembles named, role-specific agents instead of writing code: Lindy.
**Vendor-built agent teams (2 of 9 [support agents](/category/support-agents), plus one [coding agent](/category/coding-agents))** — the vendor ships multiple named, specialized agents as the product itself, with no assembly required: Crescendo, Cresta and Factory.
**Parallel fan-out** — one job runs across many agent instances at once instead of one agent working straight through: Devin, Antigravity CLI, Muse Code, Augment Code and Goose among the [coding agents](/category/coding-agents), and Manus, GPT Researcher and GC AI among the [research agents](/category/research-agents).
**Voice handoff mid-call (3 of 12 [voice agents](/category/voice-agents))** — a call starts with a triage agent and transfers to a specialist with context intact, all within one phone call: LiveKit Agents, Cartesia and Deepgram Voice Agent API.
Genspark is the sharpest exclusion on this page. Its Super Agent orchestrates nine specialized LLMs and cross-checks their output against each other, which is multi-model rather than multi-agent: one agent consulting several models is not several agents dividing a job.
Several tools sit close to the line and are deliberately out, and for two of them the distinction is worth stating plainly: being callable by an agent, or building one, is not coordinating several. Dify publishes an MCP tool that other agents can call. Replit Agent can build other agents and automations. OpenCode runs several agent sessions in parallel on the same project but names no mechanism for them to coordinate or to divide one job between them. Junie and Gemini Code Assist name no second agent anywhere in their listings.
Why this collection
A tool earns a place here only if its own Published, sourced listing documents a genuine multi-agent capability — composing multiple agents yourself, assembling named agents in a no-code builder, shipping several specialized agents as one product, fanning a single task out across parallel agent instances, or handing a live conversation off between agents mid-call — not a generic "agentic" or "automation" claim with no second agent named anywhere. Verified directly against each listing's own `whatIs`/`whatItDoes`/`keyFeatures` text, sorted into the shapes documented in [What 'Multi-Agent' Actually Means](/blog/what-multi-agent-actually-means-2026/) — not inferred from the `multi-agent` tag alone. Direction is part of the test: a tool that is merely callable BY an agent, or that builds agents for someone else to run, is not itself coordinating several, and neither earns a place here.
24 agents in this collection
Best for Engineering teams building a bespoke, production-grade agent who want the largest integration ecosystem plus real stateful control.
Ships LangGraph, the graph-runtime multi-agent control layer documented as a separate entry on this page, and bundles the wider LangChain surface (LangSmith, content blocks) used to wire coordinated agents in production.
Best for Engineering teams building production agents who need cyclical graphs, durable state and replayable runs
The stateful, graph-runtime multi-agent control layer where you design and wire multi-agent workflows as a cyclical graph — the explicit runtime piece of the LangChain surface for coordinated agents in production.
Best for Developers who want to stand up a role-based, collaborating multi-agent system quickly in Python, with a stable MIT-licensed core and a hosted build and runtime for production.
Role-based multi-agent crews where you assign named agents a role, goal and backstory and let them collaborate on a shared task — the canonical Python example of a multi-agent framework.
Best for .NET or Python teams already invested in Azure or Microsoft Foundry who want one supported agent SDK, an optional managed hosting path, and a documented migration off AutoGen or Semantic Kernel.
Microsoft's unified, MIT-licensed agent SDK for .NET and Python: the supported successor to AutoGen and Semantic Kernel, with multi-agent orchestration patterns including group chat, sequential and concurrent.
Best for Python or .NET developers and researchers exploring multi-agent orchestration who want a mature, free, self-hosted framework.
Microsoft's original multi-agent framework built around GroupChat and UserProxy patterns where agents converse to solve a task — in maintenance mode, succeeded by Microsoft Agent Framework.
Best for Teams that want Google's own official, actively-developed agent SDK, especially polyglot teams (Python/Go/Java/TypeScript/Kotlin) or anyone building multi-agent systems that need to interoperate across vendors via the A2A protocol.
Google's open-source, code-first agent SDK with five language runtimes and native A2A support — multi-agent orchestration via agents-as-tools or sub-agents handed off between agent contexts.
Best for Python developers who want an Apache 2.0 SDK with memory, knowledge, learning, guardrails and a Control Plane they run inside their own cloud account, and who are comfortable running Docker or a cloud deploy template themselves.
Apache 2.0 Python agent SDK and AgentOS runtime with built-in multi-agent teams, handoffs and a Control Plane for routing messages between agents — runnable locally or hosted on a paid tier.
Best for Developers already building on Claude (or Claude Code) who want its exact agent loop, tools and context management as a callable library, and are comfortable being single-vendor on Anthropic in exchange for the tightest possible integration with Claude Code's own capabilities.
Anthropic's own agent-building library for Python and TypeScript with explicit multi-agent sub-agent primitives: agents can spawn and delegate to other agents that own their own context.
Best for Engineers building RAG-heavy or document-heavy agents who want an MIT-licensed OSS core with an optional managed parsing and extraction platform on top.
AgentWorkflow hands control between a declared set of agents automatically, and sub-agents can be exposed to an orchestrator agent as tools — multi-agent by design, beyond the retrieval framing the name suggests.
Best for Teams that want an AI teammate embedded in Slack and a no-code app builder under the same plan, with usage that pauses at the credit cap rather than overage-billing.
A no-code builder where you assemble named, role-specific AI agents in plain English — Lindy ships the abstraction of a multi-agent team to a non-technical operator who never touches code.
Best for Mid-market and enterprise support teams that want a fully outsourced, compliance-ready AI-native CX operation, including live channels, without staffing or building one in-house.
Vendor ships multiple named, specialized AI agents as the product itself — fully-managed customer-experience AI, no user orchestration required, billed per resolved outcome.
Best for Large enterprises running high-volume
Vendor ships an autonomous AI Agent, real-time human Agent Assist and Conversation Intelligence on one contact-center platform — multi-agent by default on a single product.
Best for Engineering organizations that want AI agents handling more of the SDLC than code generation alone (triage, testing, release and documentation) and are willing to pay enterprise pricing for deployment flexibility and compliance controls.
Agent-native coding platform coordinating specialized Droids across the full software development lifecycle — the vendor ships a coordinated team of agents rather than one general coding agent.
Best for Engineering teams offloading well-scoped migration tasks without adding headcount
Coding agent that fans out across multiple sandboxes in parallel to tackle sub-tasks of a single job — true multi-agent execution rather than just multi-session in one process.
Best for Developers already inside the Google ecosystem (Google AI subscription, GCP billing) who want a free terminal agent with Gemini, Claude and gpt-oss model access, especially anyone migrating off a now-sunset Gemini CLI workflow.
Google's terminal coding agent that runs parallel agent sessions across a task — Google's own example of multi-agent coding execution on a single user-supplied job.
Best for Cost-sensitive teams willing to trade code privacy for Contributor tier's roughly 12x cheaper input and 21x cheaper output, or anyone with a large multi-part refactor that benefits from several sub-agents working in parallel isolated worktrees.
Meta's terminal coding agent that fans out parallel agents on the same problem and cross-checks their output — multi-model plus multi-agent execution pattern on a single task.
Best for Engineering leaders at organizations who want AI agents running full SDLC loops end to end, with humans at the quality gates, rather than inline completion assistants.
AI agent platform that runs multiple specialized agents in parallel across code review, ticket-to-PR delivery, security remediation and incident response — many specialized agents, one product.
Best for Devs who want an open source coding agent that reuses an existing Claude or ChatGPT plan
Open-source AI coding agent that runs multiple parallel sessions, with 70+ MCP extensions to coordinate agents across a desktop, CLI or SDK — Linux Foundation project.
Best for Individuals and teams who want to offload whole multi-step tasks (research, simple builds, data work) to an autonomous agent rather than prompt a chatbot turn by turn.
Autonomous general AI agent that plans a multi-step task and fans out across parallel sub-agents in its own cloud computer — multi-agent execution of a single user-supplied job.
Best for Developers and AI engineers who want a free, self-hosted research agent they can embed via Python, REST or MCP and run against their own LLM and search keys.
Open-source research agent that plans a question, fans out across many parallel agents to gather sources, and writes a cited report — multi-agent research, not just multi-source aggregation.
Best for In-house legal teams and corporate counsel who need fast, cited research across US case law, statutes and regulations without engaging outside counsel for a first pass.
Legal research agent that runs parallel, cross-checked research across a corpus of 13M+ US case-law opinions, statutes and regulations — multi-agent research pipeline for in-house legal teams.
Best for Engineering teams that want to own their voice-agent stack end to end, self-hosted or on LiveKit Cloud, without being locked into one vendor's API.
Apache 2.0 voice-agent framework with explicit handoff primitives: a call can transfer from a triage agent to a specialist agent with context intact, mid-call.
Best for Engineering teams that want an open-source voice-agent SDK with real LLM choice, built on speech models several rival platforms in this category already license as a component.
Line, the voice-agent SDK, supports multi-agent handoffs via its session API — a call can transfer between specialized agents mid-conversation while preserving context.
Best for Engineering teams that want a managed speech pipeline with real LLM-provider choice, and regulated enterprises that need a self-hosted or VPC deployment option a typical closed-source voice-agent vendor does not offer.
The Voice Agent API runs over a single WebSocket but supports handoff between agent configurations (LLM plus functions) mid-call, with context preserved across the transfer.
