Take CrewAI, since it's the comparison we hear most. CrewAI is a free, MIT-licensed Python framework for orchestrating teams of agents, plus a managed enterprise platform layered on top — visual studio, deployment, tracing, guardrails, enterprise security. It answers the question "how do my agents work together on tasks?" It answers it well, which is why its customer list reaches deep into the Fortune 500.
Swarm answers a different question: "how do my agents handle money?" Credentials, budgets, screened payments, a treasury with statements — and eventually the legal entity the whole operation lives inside. If you want the org-chart version: CrewAI is management and HR. We're the CFO's office and the bank. Different floors of the same building, and nobody's confused about whether a company needs both.
§ 01The telling detail is in the pricing page
Here's the observation that convinced us this boundary is real and durable: on every CrewAI tier, as of this writing, LLM token costs are pass-through. You bring your own provider keys. Free tier, enterprise tier, doesn't matter — the money that flows to model providers never routes through the framework company.
That's not an oversight. That's a boundary drawn on purpose. They stopped exactly where the money starts.
Which means every CrewAI deployment in the wild has exactly the problem our homepage describes: raw provider keys distributed across a crew of agents, all spending against someone's account, with no per-agent financial identity. No agent-level budget. No per-agent audit of spend. One key, every agent, and a single invoice at month-end that can't tell you which crew member did what.
A CrewAI crew running on scoped KeyKeeper keys instead of raw provider keys isn't a competing deployment. It's a better CrewAI deployment.
THE WHOLE ARGUMENT, ONE SENTENCE
And this generalizes. OpenClaw is our beachhead community, but the product is framework-agnostic by construction: anything that can speak an OpenAI-compatible API can hold a Swarm credential. CrewAI crews, LangGraph graphs, AutoGen teams, ElizaOS characters, a bare curl loop on a cron job — from the treasury's point of view they're all the same thing. Agents that spend.
§ 02Where the overlap gets real enough to watch
There is one contested strip. CrewAI's enterprise pitch leans on governance — compliance guardrails, role-based access, full activity logs — and "control your agents" is a message both of us send. The division underneath it:
| Question | Framework governance | Swarm |
|---|---|---|
| Governs | What agents may do | What agents may spend and earn |
| Territory | Inside the company's walls | The external economy — payments, invoices, counterparties |
| Failure it prevents | The agent does the wrong task | The agent drains the wrong account |
| Audit artifact | Activity trace | A statement your accountant can file from |
Different customers, different fears. The buyer of framework governance worries about agents misbehaving inside a workflow. Our customer worries about an agent with spending power meeting the real economy — and about being able to prove, afterward, exactly what it did with the money.
§ 03The collision that hasn't happened, and why we think it won't
The scenario that would actually put us in competition is a framework company adding native wallets and billing. Watch their revealed preference instead: the leading frameworks have had years, enormous adoption, and every opportunity to move toward the money — and they've moved the other way. Toward on-prem deployment. Toward enterprise security attestations. Toward, in at least one case, quietly removing prices from their own pricing page.
Framework companies run from money-transmission liability. It's the same de-risking reflex you see in banks that wouldn't run cash-to-Bitcoin machines and payment processors that shed anything that smells like regulatory surface. We've watched that reflex operate for a decade — Byte Federal exists because of it. Handling money in a regulated, examinable way is a burden most software companies rationally refuse.
We didn't refuse it. It's the family business.
If you run CrewAI, LangGraph, AutoGen, or anything OpenAI-compatible: keep your framework. Point its model calls at a Swarm credential instead of a raw provider key, and your crew inherits per-agent budgets, spend attribution, kill switches, and a monthly statement — without changing a line of orchestration logic.
If you're a framework maintainer reading this: the treasury layer under your users' deployments is a partnership conversation, not a rivalry. Our door is open.
So no versus page. The compare row we'll eventually add to the site is an integration strip: works with your framework — OpenClaw, CrewAI, LangGraph, AutoGen, ElizaOS, raw API. Frameworks are the labor market. We're the payroll department, the controller, and the bank branch. Buildings need floors.