Connect client

Use AgentFlowClient when your TypeScript application needs to call a running AgentFlow API.

The client does not import Python code. It sends HTTP requests to the API server you started with agentflow api.

Keep the API server running from Expose with API:

agentflow api --host 127.0.0.1 --port 8000

Install the client

In your TypeScript project:

npm install @10xscale/agentflow-client

Call the graph

Create call-agentflow.ts:

import { AgentFlowClient, Message } from "@10xscale/agentflow-client";

const client = new AgentFlowClient({
  baseUrl: "http://127.0.0.1:8000",
});

const result = await client.invoke(
  [Message.text_message("Hello from TypeScript.")],
  {
    config: {
      thread_id: "golden-path-client",
    },
    response_granularity: "low",
  },
);

console.log(result.messages.at(-1)?.text());

Run it with the TypeScript runner used by your project. For example, if your project uses tsx:

npx tsx call-agentflow.ts

Expected output:

AgentFlow API received: Hello from TypeScript.

What this call does

AgentFlowClient sends messages to /v1/graph/invoke. The Message.text_message helper creates the same message shape used by the Python API.

The config.thread_id value identifies the conversation thread for this request. Use a stable ID when you want later calls to continue the same thread.

Next step

Open the hosted playground with Open Playground.