Integration with TRPC

[promentol]

Thank you for this library and for recent feature of Agents

I realized tools for agents, they use zod schema as an arguments definition, then my question is, is it possible to use TRPC methods as a tool?

[CahidArda]

Hi there, it seems to be possible.

To test it, I have defined a checkDatabase procedure to use as a tool:

import { z } from 'zod';
import { router, publicProcedure } from './trpc';

export const appRouter = router({
  checkDatabase: publicProcedure
    .input(z.object({ id: z.string() }))
    .query(async ({ input }) => {
      // Mock database check
      const mockData = {
        '123': { name: 'John Doe', age: 30 },
        '456': { name: 'Jane Smith', age: 25 },
      };
      return mockData[input.id] || null; // Return null if not found
    }),
});

// Export type definition of API
export type AppRouter = typeof appRouter;

A tool can be defined like this and passed to the agent:

const trpc = createTRPCProxyClient<AppRouter>({
  links: [
    httpBatchLink({
      url: `http://localhost:3001/api/trpc`, // Adjust the URL accordingly
      fetch: async (input, init) => {
        return fetch(input, init)
      }
    }),
  ],
});

import { WorkflowTool } from '@upstash/workflow'
import { z } from 'zod'
import * as mathjs from 'mathjs'

const tool = new WorkflowTool({
  description:
    'A tool for evaluating mathematical expressions. ' +
    'Example expressions: ' +
    "'1.2 * (2 + 4.5)', '12.7 cm to inch', 'sin(45 deg) ^ 2'.",
  schema: z.object({ expression: z.string() }),
  invoke: async ({ expression }) => trpc.checkDatabase.query({ id }),
})

This should work. It can be improved further though. In this version, the checkDatabase will run inside context.run. However, since the query is a fetch request, using context.call is more appropriate.

To use context.call, we could overwrite the fetch method in createTRPCProxyClient like we did in ai sdk.

If you are interested, we can share a more complete example.