AI Agent Controller

This is a LangChain v1 ReAct agent intended as a starting point for building AI-powered features. It supports tool execution, chat session memory, and real-time streaming to the frontend using Server-Sent Events (SSE).

Getting Started

1. Define the agent's role

Edit the systemPrompt in createAIAgent() to describe what the agent does and which tools it can use.

systemPrompt: `You are a helpful [... e.g. travel, personal assistant, exam grading] agent.

Your responsibilities:
1. [YOUR_RESPONSIBILITY_1]
2. [YOUR_RESPONSIBILITY_2]
3. [YOUR_RESPONSIBILITY_3]

Available tools:
[LIST_YOUR_TOOLS_HERE]`

2. Replace the tools

Add tools specific to your project by replacing the existing tools in the tools array inside createAIAgent(). The existing tool functions can be removed.

Tools follow this structure and use a Zod schema for input validation:

const myTool = tool(
  async ({ input }, config) => {
    config.writer?.({ message: 'Calling my service...' });

    // Call your API or database
    const result = await callYourAPI(input);

    return JSON.stringify(result);
  },
  {
    name: 'my_tool',
    description: 'Does something specific',
    schema: z.object({
      input: z.string().describe('The input'),
    }),
  },
);