← Back to research
·9 min read·opensource

Swarms

Swarms is an Apache-2.0 Python framework for multi-agent orchestration with 40+ swarm architectures (sequential, concurrent, hierarchical, graph). 6.8K+ GitHub stars, a metered cloud API, a Solana token — and persistent community controversy around its creator.

https://swarms.ai ↗
agent-company-platformsagent-frameworksopen-sourcepython

Key takeaways

  • 6.8K+ GitHub stars and active commits as of June 2026, but star growth is modest for a project that bills itself "Enterprise-Grade Production-Ready" — and community trust is the project's biggest open question
  • Broadest architecture catalog in the space: 40+ orchestration patterns (SequentialWorkflow, ConcurrentWorkflow, HierarchicalSwarm, GraphWorkflow, MixtureOfAgents, ForestSwarm) versus the one or two patterns most frameworks ship
  • Monetizes through a metered cloud API ($6.50/1M input tokens, $18.50/1M output, $0.01 per agent), a marketplace, and a $SWARMS Solana token rather than venture funding

FAQ

What is Swarms?

Swarms is an open-source (Apache-2.0) Python framework from Kye Gomez for orchestrating multiple AI agents using composable patterns — sequential, concurrent, hierarchical, and graph-based — plus a hosted API and agent marketplace.

How much does Swarms cost?

The framework is free and self-hosted with your own model keys. The hosted Swarms API meters usage at $6.50 per 1M input tokens and $18.50 per 1M output tokens, plus $0.01 per agent and per-call tool fees.

Is Swarms an "AI agent company" platform?

Not strictly — it is a general multi-agent orchestration framework. Its hierarchical patterns (a director agent coordinating worker agents) are what people use to build company-like agent structures, which is why it appears in this category.

How is Swarms different from CrewAI?

Both are Python multi-agent frameworks; CrewAI centers on a single role-based "crew" abstraction with strong commercial backing, while Swarms ships a much larger catalog of orchestration architectures but carries significant community-trust baggage around its maintainer.

Executive Summary

Swarms is an open-source Python framework for multi-agent orchestration, created by Kye Gomez and billed as "The Enterprise-Grade Production-Ready Multi-Agent Orchestration Framework."[1] Its distinguishing feature is breadth: where most frameworks ship one or two coordination patterns, Swarms documents 40+ composable architectures — SequentialWorkflow, ConcurrentWorkflow, HierarchicalSwarm, GraphWorkflow, MixtureOfAgents, GroupChat, MajorityVoting, ForestSwarm, and more.[2] It is not a company-of-agents platform per se; it earns a place in this category because its hierarchical patterns — a director agent delegating to specialized workers — are the building blocks people use to assemble company-like agent structures.

The repository holds 6.8K+ stars and 949 forks as of June 2026, with commits landing days before this writing, so the project is unambiguously active.[1] The complication is reputational. Kye Gomez has drawn sustained community criticism — Hacker News threads alleging paper-mimicking repos that don't run, issues closed without replies, and a widely circulated legal threat against OpenAI over the "swarm" name.[3][4] Swarms also funds itself unconventionally: no disclosed venture round, but a metered cloud API, an agent marketplace, and a $SWARMS token on Solana.[5][6]

AttributeDetail
CreatorKye Gomez / Swarms.ai
Repo createdMay 11, 2023 [1]
FundingNo disclosed venture funding; $SWARMS Solana token and API/marketplace revenue [6]
GitHub Stars6.8K+ (June 2026) [1]
LicenseApache-2.0 [1]
LanguagePython [1]

Product Overview

Swarms is a pip install-able framework: you define agents (model, system prompt, tools, memory), then compose them into a swarm architecture that controls how they communicate and divide work. The docs claim enterprise features — observability, multi-model provider support, MCP/AOP protocol support, an agent registry, and backwards compatibility with LangChain, AutoGen, and CrewAI agents — alongside a marketplace for sharing production-ready prompts and agents.[2]

Key Capabilities

CapabilityDescription
Architecture catalog40+ patterns: sequential, concurrent, hierarchical, graph (DAG), mixture-of-agents, group chat, majority voting, debate, council-as-judge [2]
Hierarchical orchestrationHierarchicalSwarm director coordinates worker agents; HybridHierarchicalClusterSwarm routes tasks to sub-swarms — the company-like pattern [2]
Auto-constructionAutoSwarmBuilder decomposes a task and assembles a swarm automatically [2]
SchedulingCronJob runs agents on cron-style intervals [2]
Protocol supportMCP, X402, AOP (deploy agents as MCP-server tools) [2]
Framework interopClaimed backwards compatibility with LangChain, AutoGen, CrewAI [2]

Product Surfaces

SurfaceDescriptionAvailability
Python frameworkOpen-source library, self-hosted, BYO model keysFree, Apache-2.0 [1]
Swarms APIHosted agent/swarm execution, metered per token and per agentUsage-based [5]
MarketplaceDiscover and share prompts and agentsLive [2]
$SWARMS tokenSolana token for governance and ecosystem incentivesLive [6]

Technical Architecture

Swarms is a single Python package installed via pip, uv, or poetry.[2]

pip install swarms

Agents wrap any supported model provider; swarm classes are the orchestration layer. Graph-based workflows (GraphWorkflow) handle DAG execution with parallelism; embedding-based routers (AgentRouter, ForestSwarm) match tasks to agents by semantic similarity.[2]

Key Technical Details

DetailValue
DeploymentSelf-hosted Python library, or hosted Swarms API [2]
Model(s)Multi-provider; model-agnostic agent wrapper [2]
IntegrationsMCP, AOP, LangChain/AutoGen/CrewAI compatibility, cron scheduling [2]
Open SourceYes — Apache-2.0 [1]

Strengths

  • Unmatched pattern breadth — 40+ documented orchestration architectures, from simple pipelines to hierarchical clusters and debate/judge ensembles; most rivals ship a fraction of that.[2]
  • Genuinely active — pushed within days of this writing, with 949 forks and a long release history.[1]
  • Permissive license — Apache-2.0, self-hosted, no account required for the framework itself.[1]
  • Hierarchy as a first-class citizen — director/worker swarms, sub-swarm routing, and auto-built swarms map naturally onto org-chart-style agent companies.[2]
  • Transparent API pricing — published unified per-token and per-agent rates rather than opaque enterprise quotes.[5]

Cautions

  • Maintainer controversy is the headline risk. Hacker News commenters allege a pattern of repos mimicking just-published papers that "never actually run," issues closed without replies, and stars accumulated via social-media promotion rather than working code.[3]
  • The OpenAI legal-threat episode. When OpenAI released its own "Swarm" framework, an issue titled "Notorious namesquatter is threatening legal action" alleged Gomez demanded $10M over the name despite holding no trademark; HN commenters considered the claim baseless given decades of prior "swarm" usage.[4][3]
  • Public disputes with other founders — ai16z's Shaw publicly called the project's code into question; Gomez disputed the claims as evidence-free, but the episode reinforced the trust gap.[7]
  • Crypto-token financing — a $SWARMS Solana token funds the ecosystem in lieu of disclosed venture backing; token-driven incentives can attract speculation rather than production users.[6]
  • "Enterprise-grade" claims outpace evidence — the docs advertise a 99.9%+ uptime guarantee, but there are no published reference customers to anchor the positioning.[2]
  • Modest traction for its age — 6.8K+ stars after three years is respectable but far from category leaders, and the framework predates (and competes with) much better-capitalized rivals.[1]

Pricing & Licensing

TierPriceIncludes
Framework (self-hosted)FreeFull library, BYO model keys, Apache-2.0 [1]
Swarms API$6.50 / 1M input tokens, $18.50 / 1M output tokensHosted execution, unified rate across endpoints [5]
API add-ons$0.01 per agent; $0.25 per image; $0.10 per MCP call; $0.04 per search; $0.15 per scrapeMetered tool and agent surcharges [5]

Licensing model: Apache-2.0 open core; hosted API and marketplace are the commercial layer, with a $SWARMS token funding the broader ecosystem.[1][6]

Hidden costs: self-hosting means your own model-provider bills; the API's per-agent and per-tool surcharges compound quickly in large swarms; evaluating code quality yourself is effectively mandatory given the community-trust history.[3]

Competitive Positioning

Direct Competitors

CompetitorDifferentiation
CrewAISingle role-based "crew" abstraction, commercial backing, larger community; Swarms counters with a far broader architecture catalog [2]
LangChain (LangGraph)Graph-first orchestration with a massive ecosystem; Swarms claims backwards compatibility with it rather than head-on replacement [2]
PaperclipPurpose-built company/org-chart platform with budgets and governance; Swarms gives you the hierarchical primitives to build that yourself, without the employee metaphor

When to Choose Swarms Over Alternatives

  • You want to experiment across many orchestration topologies (debate, voting, hierarchy, DAG) without switching frameworks.[2]
  • You need Apache-2.0 licensing and full self-hosting with no account.[1]
  • You want hierarchical director/worker primitives rather than a prescriptive company platform.

Ideal Customer Profile

Best fit:

  • Python developers prototyping multi-agent topologies who value pattern variety over polish
  • Teams building custom hierarchical agent structures who want primitives, not an opinionated org-chart product
  • Builders already in the Solana/agent-token ecosystem where $SWARMS circulates [6]

Poor fit:

  • Enterprises that vet vendor reputation — the maintainer controversy will not survive procurement review [3]
  • Teams wanting built-in governance, budgets, and human-approval flows out of the box
  • Anyone needing a framework whose claims are backed by published reference customers

Viability Assessment

DimensionAssessment
Financial HealthOpaque — no disclosed venture funding; revenue from metered API plus a volatile crypto token [6]
Market PositionNiche — 6.8K+ stars after three years, well behind CrewAI and LangChain ecosystems [1]
Innovation PaceHigh on surface area — continuous commits and a steadily expanding architecture catalog [1][2]
Community/EcosystemDamaged — repeated public controversies over code quality, namesquatting, and legal threats [3][4]
Long-term OutlookUncertain — survival depends on converting breadth into trust, against better-funded competitors

Swarms is a paradox: one of the most architecturally ambitious open-source agent frameworks, actively maintained for three years, yet carrying enough reputational baggage that many teams will rule it out before evaluating the code. The Apache-2.0 license caps the downside — the patterns are inspectable and forkable — but the token-financed model and contested maintainer history make it a research dependency, not an enterprise one.

Bottom Line

Swarms delivers the widest catalog of multi-agent orchestration patterns in the open-source field, and its hierarchical swarms are legitimate raw material for company-like agent structures — but it is a framework you adopt with eyes open, because the project's community-trust deficit is as well-documented as its feature list.[2][3]

Recommended for: Python tinkerers exploring orchestration topologies; teams comfortable auditing Apache-2.0 code and bringing their own keys.

Not recommended for: enterprises with vendor-risk review; teams wanting an opinionated, governed agent-company platform like Paperclip.

Outlook: active development continues, but absent a reputational turnaround or marquee production users, Swarms likely remains a pattern library people learn from more than a platform they bet on.

Research by Ry Walker Research • methodology

Sources

Share on XShare on LinkedIn