Agency Swarm
What is Agency Swarm?
Agency Swarm is an open-source framework for developers who need to coordinate multi-agent workflows with direct control over roles, prompts, and custom Python tools. It includes Customizable Agent Roles, Full Control Over Prompts, Error Correction, Efficient Communication, Custom Tools, CodeInterpreter, Deployable in Production, and Documentation Index, with FastAPI integration and an llms.txt documentation index. It is used by VRSEN/agency-swarm and has 4,587 GitHub stars.
Last verifiedHow we evaluate
At a glance
- Agency Swarm is best for developers who need to orchestrate multi-agent workflows with direct control over roles and prompts.
- Yes — The page points to a complete documentation index for discovering the framework's available pages and capabilities.
What it does well
Agency Swarm is built directly on top of OpenAI's own Agents SDK and Responses API rather than wrapping a third framework, and tools are defined as Pydantic models, which gives automatic argument validation instead of the looser type-hint-only checking some competitors use. Communication between agents is directional and explicit (ceo > dev style flows) rather than an open broadcast, which makes larger agent graphs easier to reason about. The project has shipped continuously since its first PyPI release in January 2024 (agency_swarm 0.1.0) through v1.11.0 in August 2026, and GitHub shows commits as recently as 30 September 2026 — this is not a stalled project (https://pypi.org/pypi/agency-swarm/json, https://api.github.com/repos/VRSEN/agency-swarm).
Built around OpenAI's SDK, not model-agnostic by default
The framework's core dependency is openai-agents, pinned to an exact version (0.18.1 as of the 1.11.0 release), and its default model as of that release is OpenAI's gpt-5.6-luna. Anthropic, Google Gemini, xAI Grok, Azure OpenAI and OpenRouter models are supported, but only via a LiteLLM router layer rather than as first-class citizens (https://github.com/VRSEN/agency-swarm/releases/tag/v1.11.0, https://pypi.org/pypi/agency-swarm/json). It also requires Python 3.12 or newer, which rules out teams on older Python runtimes. A team standardized on OpenAI gets a tightly integrated experience; a team wanting equal-footing support for other model providers is working through an extra abstraction layer.
Two products share this name — know which one you're looking at
Agency Swarm the framework is MIT-licensed, free, and self-hosted: pip install agency-swarm. Agencii, built on top of it by the same team, is a separate hosted deployment platform with its own pricing: a free tier (25 daily credits, up to 5 agencies) and a Pro tier at $20/month (25 daily + 2,000 monthly credits, unlimited agencies), plus one-time credit refills at $30/3,000 or $100/10,000 (https://agency-swarm.ai/platform/pricing, checked 2026-10-02). None of this applies if you only use the open-source framework and self-host; it only matters if you want the vendor's managed runtime instead.
No independent reviews found
Searches turned up tool-directory listings (PyPI, DeepWiki, aggregator sites) that restate the vendor's own description, plus the vendor's own comparison page claiming advantages over AutoGen and CrewAI on reliability, flexibility and deployability — a vendor claim, not an independent benchmark (https://agency-swarm.ai/welcome/ai-agency-vs-other-frameworks). No third-party technical review, benchmark, or critical writeup was found as of this check.
Security and compliance not yet formalized
The vendor's platform page lists SOC 2 compliance, SSO, and audit capabilities under a 'Future Roadmap' section, explicitly marked as not yet built (https://agency-swarm.ai/platform/overview). There is no published security/trust page for either the framework or the Agencii platform as of this check.
Frequently asked questions
What is Agency Swarm?
Agency Swarm is an open-source framework for developers who need to coordinate multi-agent workflows with direct control over roles, prompts, and custom Python tools. It includes Customizable Agent Roles, Full Control Over Prompts, Error Correction, Efficient Communication, Custom Tools, and CodeInterpreter, with FastAPI integration and an llms.txt documentation index. It is used by VRSEN/agency-swarm and has 4,587 GitHub stars.
What is Agency Swarm used for? Who is it for?
Agency Swarm is used for Customizable Agent Roles, Full Control Over Prompts, and Error Correction. It's built for AI application developers, Python engineers, and Teams shipping agent-backed APIs.
Does Agency Swarm have an API and what does it integrate with?
The page points to a complete documentation index for discovering the framework's available pages and capabilities. It integrates with OpenAI, FastAPI, GitHub.
Editor's read
Check whether your workflow needs FastAPI-based deployment or only local orchestration. The docs show production guidance and self-hosting, so confirm that your team actually plans to serve the agency as an API before adopting it.
Explore other Agent Frameworks
