Swarms API Documentation - Build AI Agents & Multi-Agent Systems

The Swarms API is a hosted, model-agnostic multi-agent REST API. Unlike Python-only orchestration frameworks, you call it over HTTP from any language, deploy nothing locally, and pay only for what you run. It handles all agent scheduling, context passing, parallel execution, and retries on its side.

Capability	LangGraph	CrewAI	AutoGen	LangChain	Swarms API
Deployment	Self-hosted	Self-hosted	Self-hosted	Self-hosted	Fully hosted
Language	Python	Python	Python	Python / JS	Any (REST)
Multi-agent	Graph DAG	Sequential / Hierarchical	GroupChat	Chains / Agents	10+ architectures
State management	Explicit state graph	Task-based	Conversation history	Memory modules	Handled by API
Parallelism	Async / compiled graph	Limited	GroupChat turns	Limited	Native parallel execution
Model support	OpenAI, Anthropic, etc.	OpenAI, Anthropic, etc.	OpenAI, Anthropic, etc.	100+ via LiteLLM	300+ via unified endpoint
Pricing	Infra cost + model cost	Infra cost + model cost	Infra cost + model cost	Infra cost + model cost	Per-token + per-agent

Concept Mapping

The table below maps every major concept from each framework to its Swarms API equivalent.

Agents

Framework	Their Concept	Swarms API Equivalent
LangGraph	`node` (a callable or Runnable)	`agent` object in `agents` array
CrewAI	`Agent(role, goal, backstory)`	`agent` with `agent_name` + `system_prompt`
AutoGen	`AssistantAgent` / `ConversableAgent`	`agent` with `system_prompt`
LangChain	`AgentExecutor` / `LLMChain`	Single agent completion or `SequentialWorkflow`

Workflows & Orchestration

Framework	Their Concept	Swarms API Equivalent
LangGraph	`StateGraph` with conditional edges	`GraphWorkflow` with `edges`, `entry_points`, `end_points`
CrewAI	`Crew(process=Process.sequential)`	`SequentialWorkflow`
CrewAI	`Crew(process=Process.hierarchical)`	`HierarchicalSwarm`
AutoGen	`GroupChat` + `GroupChatManager`	`GroupChat` swarm type
LangChain	`SequentialChain`	`SequentialWorkflow`
LangChain	Parallel `RunnableParallel`	`ConcurrentWorkflow`

Tasks & Prompts

Framework	Their Concept	Swarms API Equivalent
LangGraph	State dict passed between nodes	Agent output is appended to context automatically
CrewAI	`Task(description, agent, expected_output)`	Top-level `task` string; agent role defined in `system_prompt`
AutoGen	`initiate_chat(message)`	Top-level `task` string
LangChain	`PromptTemplate` + `chain.invoke(input)`	`system_prompt` + `task`

Tools

Framework	Their Concept	Swarms API Equivalent
LangGraph	`ToolNode` / `bind_tools`	`tools_enabled` array on the request (e.g. `"auto_search"`, `"web_scraper"`), or custom function schemas via `tools_list_dictionary`
CrewAI	`@tool` decorated functions	`tools_list_dictionary` (OpenAI-style function schemas) on the agent
AutoGen	`register_function`	`tools_list_dictionary` (OpenAI-style function schemas) on the agent
LangChain	`Tool` / `BaseTool`	`tools_list_dictionary` (OpenAI-style function schemas) on the agent

Architecture Selection Guide

Once you know what you were building in your old framework, use this table to pick the right Swarms workflow:

What you were building	Best Swarms architecture
Linear pipeline (A → B → C)	`SequentialWorkflow`
Parallel fan-out (all agents same task)	`ConcurrentWorkflow`
Graph with mixed parallel + sequential	`GraphWorkflow`
Hierarchical with a manager agent	`HierarchicalSwarm`
Route tasks to the right specialist	`MultiAgentRouter`
Multiple experts debate and consensus	`MajorityVoting`
Open-ended group discussion	`GroupChat`
Compare N approaches to same problem	`MixtureOfAgents`
Same tasks × multiple agents (grid)	`BatchedGridWorkflow`

Universal Migration Checklist

Regardless of which framework you are migrating from, follow these steps:

Get your API key at swarms.world/platform/api-keys
Set the environment variable: export SWARMS_API_KEY="your-key"
Install the HTTP client of your choice (requests, httpx, fetch, axios, etc.)
Map each agent in your old workflow to an agent object with agent_name, system_prompt, and model_name
Map the topology — sequential chain, parallel fan-out, or directed graph
Replace task inputs — the top-level task field replaces all invoke(), kickoff(), and initiate_chat() calls
Remove local infrastructure — no more Python environments, LLM SDK imports, or API key plumbing per-library

Migration Guides

Choose your current framework:

Quick Start After Migration

Every Swarms API call follows the same pattern regardless of workflow type:

import os
import requests

response = requests.post(
    "https://api.swarms.world/v1/<workflow-endpoint>/completions",
    headers={
        "x-api-key": os.environ["SWARMS_API_KEY"],
        "Content-Type": "application/json",
    },
    json={
        "name": "My Workflow",
        "description": "What this workflow does",
        "task": "The task to complete",
        "agents": [ ... ],   # your agents here
        # workflow-specific fields (edges, entry_points, etc.)
    },
    timeout=300,
)

result = response.json()

The base URL is https://api.swarms.world. See the API Reference for all available endpoints.

Getting Started

Agent Capabilities

Multi-Agent Architectures

Finance & Investing

Healthcare & Life Sciences

Legal, M&A & Diligence

Customer Operations

Defense, Gov & Supply Chain

Marketing, Content & Founder Tools

API Utilities

Migration Guides

Resources

Production

Compatibility & Benchmarks

Tutorials & Integrations

Migration Overview

Concept Mapping

Agents

Workflows & Orchestration

Tasks & Prompts

Tools

Architecture Selection Guide

Universal Migration Checklist

Migration Guides

Quick Start After Migration

​Concept Mapping

​Agents

​Workflows & Orchestration

​Tasks & Prompts

​Tools

​Architecture Selection Guide

​Universal Migration Checklist

​Migration Guides

​Quick Start After Migration

Concept Mapping

Agents

Workflows & Orchestration

Tasks & Prompts

Tools

Architecture Selection Guide

Universal Migration Checklist

Migration Guides

Quick Start After Migration