Our new open-source repository collects 100+ runnable examples for the Swarms API, covering single agents, coordinated swarms, every orchestration pattern, model integrations across GPT-5, Claude, Gemini, Grok, DeepSeek and more, plus streaming, vision, tools, and real industry use cases.
The hardest part of building a multi-agent system is rarely the idea. It is the hour you spend working out which parameters the API actually wants, how a hierarchical swarm differs from a concurrent one in practice, and whether streaming works the way you assumed it does. Documentation answers the first question. Working code answers all three.
So we published the working code. The Swarms API Examples repository is an open-source collection of more than 100 runnable scripts for the Swarms API, organized so you can find the one closest to what you are building, run it, and edit it into your own system. Every example is meant to execute as written once you have an API key. Nothing is a fragment or a sketch.
Single agents and coordinated swarms. The examples start at the beginning: one agent, one prompt, one response. From there they move through multi-agent setups where several agents share a task and hand work between each other. If you have never called the API before, the getting-started scripts are the fastest path from zero to a response.
Every orchestration pattern. Sequential workflows where agents run in order, concurrent workflows where they run at once, hierarchical swarms with a director delegating to workers, and graph workflows where you define the dependencies and let the structure decide execution order. Each pattern gets its own example, which makes the differences between them concrete rather than theoretical. Choosing an architecture is much easier when you can run all four against the same task and compare the output.
Frontier model integrations. GPT-5, Claude, Gemini, Grok, DeepSeek, Groq, and OpenRouter each have examples showing exactly how to point an agent at them. Swapping the model behind an agent is usually a one-line change, and these scripts show you which line.
Real industry use cases. Healthcare, finance, marketing, and software engineering examples built around problems those fields actually have, not generic demos. They are useful as starting points and also as evidence of what a well-scoped agent system looks like in a specific domain.
The full surface of the platform. SDK usage and raw cURL calls for anyone working outside Python. MCP examples for connecting external tools. Streaming for token-by-token output, reasoning examples for models that expose their thinking, vision examples for image input, tool-calling examples, and platform utilities for the operational work around your agents.
Clone the repository, add your API key, and run any script in it. Most people will get further faster by opening the example nearest their use case and modifying it than by reading the reference documentation front to back. Both are available, and they complement each other.
The repository is open source, so pull requests are welcome. If you build something with the API that would have saved you time as an example, it probably belongs in there.
Examples are the shortest distance between an idea and a running system. This repository exists so that the time you spend on a multi-agent project goes into the part that is specific to your problem, rather than into rediscovering how the API is shaped.
Clone it, run something, and build from there.
| Resource | Link |
|---|---|
| Swarms API Examples on GitHub | The-Swarm-Corporation/Swarms-API-Examples |
| Sign up | cloud.swarms.world |
| API keys | cloud.swarms.world/api-keys |
| Documentation | docs.swarms.ai |
| Swarms Marketplace | swarms.world |
| Twitter/X | @swarms_corp |
A day by day log of everything that shipped on the Swarms Marketplace over the last month: on-chain agent competitions with real prize pools, the tokenized agent Screener, a dedicated MCP Servers page, a self-updating public API spec, a rebuilt home page and sign-in, a full mobile pass, and a serious security hardening effort, each written in plain language so you know what changed and what it means for you.
The Swarms research team has published a full systems paper on GraphWorkflow, the graph execution engine inside the Swarms framework. Across an open benchmark suite of five topologies at 10 to 200 nodes, GraphWorkflow executes compiled agent graphs with a geometric-mean speedup of 7.0x over LangGraph, rising to 62.5x on deep chains, compiles graphs 21.6x to 31.3x faster, and completes the cold build-compile-execute path 7.9x faster. This article walks through the paper: the cost taxonomy, the compile-once architecture, the programming model comparison, the full benchmark results with figures from the paper, and how to reproduce every number yourself.
Swarms Cloud now has a private, encrypted library for your prompts and skills at cloud.swarms.world/skills. Save prompts by hand or drag and drop Anthropic-format SKILL.md files, organize them with tags and search, give every entry its own page, and store all of it encrypted with a key derived from your account, so only you can ever read it. Available to every user on every plan: Free, Pro, and Premium.