Restless vs building your own with CrewAI, LangGraph or Microsoft Agent Framework.
Agent frameworks are toolkits for developers building their own agent apps. Restless is an app you use. Here's how to tell which one you actually need.
"Why not just build it with CrewAI?" is a fair question, especially from developers. Frameworks like CrewAI, LangGraph and Microsoft Agent Framework are genuinely good, and plenty of people build impressive things with them.
The difference is simple, though: a framework is a box of parts, and Restless is a finished room. Which one you want depends on whether you're trying to build an agent system or use one.
What the frameworks are
CrewAI is a Python framework for coordinating role-playing agents. You define agents with a role, goal, backstory, model and tools, give them tasks, and group them into crews that collaborate on their own, or use flows when you want precise, event-driven control over execution and state.
LangGraph is a low-level orchestration framework for long-running, stateful agents, in Python and JavaScript. It gives you durable execution that resumes after failures, human-in-the-loop points where you can inspect and change agent state, and short- and long-term memory.
Microsoft Agent Framework is an open framework for building agents and multi-agent workflows in Python and .NET, with sequential, concurrent, handoff and group patterns. It links migration guides from both AutoGen and Semantic Kernel.
All three are MIT licensed.
When a framework is the right call
You should probably reach for a framework if:
- You're building an agent product or feature for your own customers.
- You want full control over the execution model, state and every prompt.
- Your team writes code and wants agents embedded in your existing systems.
- You need a very specific orchestration pattern no off-the-shelf app supports.
What Restless gives you instead
Restless is the thing you'd otherwise spend months building on top of a framework, aimed at running work rather than shipping an agent API:
- A workspace people actually use. Rooms, live co-edited documents, comments, a board of work and an inbox for decisions, with real accounts and permissions for your teammates.
- A company computer per company. A persistent Linux machine with files, Git, a browser and a desktop, with resource limits and lifecycle handled for you.
- Organisation that survives restarts. Goals, owners and work history live outside any one agent session.
- Authority in code. Permissions, spending limits, a credentials vault and receipts for every action taken in the outside world.
- Native agent runtimes. Codex and Claude Code harnesses, or any API model, rather than wrapping a raw model call yourself.
The hidden cost of rolling your own
Getting a demo crew working is a weekend. Getting it to survive real use is the long tail: where does state live when the process dies, who's allowed to spend money, how do humans review the actual output, what happens when a request times out after the provider might have done it, how do you stop forgotten processes eating your machine for a day. I know, because that list is basically my commit history.
None of that makes frameworks bad. It just means the framework is the start of the project, not the end of it.
Honest trade-offs
Frameworks give you far more flexibility than Restless will. You can shape every part of the system, embed it anywhere and ship it to customers. Restless is opinionated about how work is organised, and it's a development preview. If you need agents inside your own product, a framework is the right foundation.