GPT‑5 Agent (Node/TypeScript)

General-purpose workspace assistant powered by OpenAI gpt-5, built with Express and VoltAgent. It receives batched channel events via a simple HTTP endpoint and replies back to the channel using the agent-chat service through a dedicated tool.

Features

• Event-driven: HTTP endpoint POST /v1/events for incoming channel events
• Tool-gated messaging: Always replies via the sendMessageToChannel tool
• Clean agent setup: VoltAgent + Vercel AI provider targeting openai("gpt-5")
• Typed payloads: Request schemas encoded in TypeScript

Prerequisites

• Node.js 18+
• pnpm
• Required environment variables:
  • OPENAI_API_KEY: OpenAI API key for the openai("gpt-5") model
  • AGENT_API_KEY: API key for the agent-chat service
  • PORT (optional, default 7102)
  • SLACK_VERIFICATION_TOKEN (optional; only needed if you enable request verification middleware)

Create a .env file in gpt-5/:

OPENAI_API_KEY=sk-...
AGENT_API_KEY=...
PORT=7102
SLACK_VERIFICATION_TOKEN=optional-token

Install and run

pnpm i
pnpm dev      # runs src/index.ts with ts-node

# or build and run
pnpm build
pnpm start

Server listens on /:port (default 7102).

HTTP API

• POST /v1/events
  • Returns 200 OK immediately; processing continues asynchronously.
  • Body (TypeScript type IPostEventsRouteBody):

type IPostEventsRouteBody = {
  event: "channel.messages.batch";
  workspaceId: string;
  channelId: string;
  messages: Array<{
    id: string;
    content: string;
    sender_id: string;
    status: "sent";
    timestamp: number;
    type: string;
  }>;
  timestamp: number;
  context: {
    relevantWorkspaceContext: {
      memories: Array<{
        chunks: Array<{ content: string; isRelevant: boolean; score: number }>;
        summary: string;
      }>;
      time: number;
      total: number;
    } | null;
  };
};

Example request:

curl -X POST http://localhost:7102/v1/events \
  -H 'Content-Type: application/json' \
  -d '{
    "event": "channel.messages.batch",
    "workspaceId": "ws_123",
    "channelId": "ch_456",
    "messages": [
      {"id": "m1", "content": "@gpt5 Summarize this thread", "sender_id": "u1", "status": "sent", "timestamp": 1736111111, "type": "text"}
    ],
    "timestamp": 1736111112,
    "context": {"relevantWorkspaceContext": null}
  }'

Agent behavior

• Defined in src/agents/gpt5/gpt5Agent.ts with:
  • name: gpt5
  • model: openai("gpt-5")
  • Instructions emphasize that the agent must reply via the sendMessageToChannel tool.
• On incoming events, the controller:
  • Loads tools via loadTools
  • Invokes agent.generateText(JSON.stringify(req.body), { conversationId: channelId, userId: workspaceId })
  • The agent must call the sendMessageToChannel tool to send the actual response back to the channel.

Tools

• sendMessageToChannel (src/agents/gpt5/tools/sendMessageToChannel)
  • Parameters: { channelId: string; message: string }
  • Sends a message via agent-chat using agentChat.channel.message.send({ channelId, content })
  • Requires AGENT_API_KEY to be set

Optional: Request verification

A simple middleware exists at src/routes/events/middlewares/verifyRequest to check req.body.token === process.env.SLACK_VERIFICATION_TOKEN.

To enable it, update the events router:

// src/routes/events/index.ts
import { Router } from "express";
import { postEventsController } from "@/controllers/events";
import { verifyRequest } from "./middlewares";

const router = Router();
router.post("/", verifyRequest, postEventsController);
export default router;

Project structure (key files)

src/
  index.ts                          # Express server, mounts /v1/events
  agents/
    agents.ts                       # Agent registry
    gpt5/
      gpt5Agent.ts                  # GPT-5 agent definition
      constants.ts                  # Name, description, instructions
      tools/
        loadTools.ts                # Registers tools for the agent
        sendMessageToChannel/       # Tool to send channel messages via agent-chat
  controllers/
    events/post/postEventsController.ts
  routes/
    events/index.ts                 # POST /v1/events
    events/middlewares/verifyRequest
  config/
    index.ts                        # env, alias registration, exports agentChat
    agentChat/agentChat.ts          # AgentChat client (needs AGENT_API_KEY)

License

ISC (see package.json).