Use Agent200 from any agent framework
LangGraph, the OpenAI Agents SDK or your own code: Agent200 plugs in where your agent reaches the outside world, and your framework carries on as before.
2 min read
You already chose how your agent thinks. Maybe it is a LangGraph graph, maybe it runs on the OpenAI Agents SDK, maybe it is a few hundred lines of your own code that you understand completely. Agent200 does not ask you to give any of that up. It plugs into the place where your agent reaches the outside world.
Where the SDK goes
Anywhere your agent needs a service, a model or a workflow, it makes one Agent200 request. That can be:
- a node in a graph that gathers context before the next step,
- a tool your existing agent can call when a task needs external data,
- a route in your API that answers a user’s question,
- a background job your own scheduler starts every morning.
Inside that request, Agent200 runs the model you chose with the tools you allowed, executes the tool calls, and hands back the result. Around it, your framework carries on exactly as before.
A node that gathers context
Conceptual example.
async function gatherContext(state) {
const result = await agent200
.user(state.userId)
.run({
model: "openai/gpt-5.6-sol",
instructions: "Collect what the user needs for their next meeting.",
input: state.request,
tools: ["brief"],
});
return { ...state, context: result };
}The graph does not know or care that a workflow, two services and a model ran inside that step. It receives the context and moves to the next node.
Who owns what
| Your agent keeps | Agent200 provides |
|---|---|
| Orchestration and the order of steps | Connectors for supported services |
| Business logic and product decisions | Managed OAuth and per-user connections |
| When it runs, and how often | The tool-calling loop inside each request |
| Its interface and its users | Models and paid services through one API key |
Runs when you call it
Agent200 executes when your application makes a request. If a summary should arrive every Monday at nine, your scheduler starts that request at nine. That keeps timing, retries at the application level and the shape of your product in your hands, where they belong.
Switching frameworks later
Because the external capabilities live behind one SDK, they are not tied to your framework. Move from a graph to plain code, or the other way round, and the connectors, the workflows and the accounts your users already connected stay where they are.
See how this fits together on the Platform page.
