Documentation Index
Fetch the complete documentation index at: https://docs.swarms.ai/llms.txt
Use this file to discover all available pages before exploring further.
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 |
| Framework | Their Concept | Swarms API Equivalent |
|---|
| LangGraph | ToolNode / bind_tools | tools array on agent (e.g. "browser", "code_interpreter") |
| CrewAI | @tool decorated functions | tools array on agent |
| AutoGen | register_function | tools array on agent |
| LangChain | Tool / BaseTool | tools array on 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()
https://api.swarms.world. See the API Reference for all available endpoints.
