Quickstart

This example uses LangChain createAgent; Agentdock does not construct or run the model/tool loop. Install a provider integration and set its credentials.

import { Agentdock } from "@agentdock-ai/agentdock";
import { ChatOpenRouter } from "@langchain/openrouter";
import { MemorySaver } from "@langchain/langgraph";
import { createAgent, tool } from "langchain";
import { z } from "zod";

const weather = tool(async ({ city }) => ({ city, forecast: "Sunny" }), {
  name: "get_weather",
  description: "Get the current weather for a city.",
  schema: z.object({ city: z.string() }),
});

const graph = createAgent({
  model: new ChatOpenRouter({ model: "openai/gpt-4o-mini" }),
  tools: [weather],
  checkpointer: new MemorySaver(),
}).graph;

const runtime = new Agentdock(graph);

After your application authenticates and authorizes a request, pass an application-derived thread ID and graph input to the serving runtime:

await runtime.pipe(response, {
  threadId: threadIdFromAuthorizedSession,
  input: { messages: [{ role: "user", content: "Weather in Lahore?" }] },
  context: { userId: authenticatedUser.id },
});

pipe() writes Server-Sent Events to a Node response, honors backpressure, and aborts cooperative graph and tool work if the client disconnects. Use toResponse() for a Web Response, or stream() when you want to consume events without a transport.

MemorySaver is for development and single-process demonstrations. Use an appropriate LangGraph saver for durable production checkpoints, and manage its lifecycle in your application.