Migrating from Responses + OpenAI Vector Stores to my own RAG Tool with Agents: looking for feedback on tool design

API
, , ,
1

I was using this architecture:

A backend was calling OpenAI Responses API, sending a Prompt and a vectorStoreId

At that time, I had no idea how the Assistant was getting in contact with the Vector Store. I basically didn’t know (and I don’t know) what was the “tool” description and the “query” (parameter) description of the “vector-searcher-tool”, because that’s a black-box on OpenAI.

Now, I’m migrating to Agentics. I have to create my own Vector Store, and get the vectors from a Tool. I basically have an Agent, who has a Tool, who calls my own vector store. The problem I have, is that I don’t know how to describe the tool, and how to describe the “query” parameter in my own tool.

Here a code example:

import { tool } from '@openai/agents';
import { z } from 'zod';


const tool = tool({
  name: 'searcher',
  description: `What should I set here in order to simulate the VectorStore tool on OpenAI?`,
  parameters: z.object({
    query: z
      .string()
      .describe(
        'And what about here?'
      ),
  })
});

const agent = new Agent({
  name: 'agent',
  instructions: `Here is OK.`,
  tools: [tool],
});
1 Like