Convex Agents Fundamentals

Purpose

Guides you through the core patterns for setting up Convex agents, managing conversation threads, and generating LLM responses. This is the foundation upon which all other agent capabilities build.

When to Use This Skill

• Setting up your first Convex agent in a project
• Creating or continuing conversation threads with users
• Generating text responses or structured JSON objects from an LLM
• Configuring agent defaults (system prompt, chat model, embedding model)
• Building basic chat interfaces that need message history

How to Use It

1. Install and Configure

Add the agent component to your convex.config.ts:

// convex/convex.config.ts
import { defineApp } from "convex/server";
import agent from "@convex-dev/agent/convex.config";

const app = defineApp();
app.use(agent);

export default app;

Run npx convex dev to generate the required code.

2. Define Your Agent

Create an agent instance with a chat model:

// convex/agents/myAgent.ts
import { components } from "../_generated/api";
import { Agent } from "@convex-dev/agent";
import { openai } from "@ai-sdk/openai";

export const myAgent = new Agent(components.agent, {
  name: "My Assistant",
  languageModel: openai.chat("gpt-4o-mini"),
  instructions: "You are a helpful assistant.", // Optional: default system prompt
});

3. Create Threads

Create a thread for a user to manage their conversation history:

// convex/threads.ts
import { action } from "../_generated/server";
import { v } from "convex/values";
import { myAgent } from "./agents/myAgent";

export const createNewThread = action({
  args: { userId: v.string() },
  handler: async (ctx, { userId }) => {
    const { thread } = await myAgent.createThread(ctx, {
      userId,
      title: "New Conversation",
    });
    return { threadId: thread.getMetadata().threadId };
  },
});

4. Generate Responses

Generate text or structured responses in a thread:

// convex/generation.ts
export const generateReply = action({
  args: { threadId: v.string(), prompt: v.string() },
  handler: async (ctx, { threadId, prompt }) => {
    const { thread } = await myAgent.continueThread(ctx, { threadId });
    const result = await thread.generateText({ prompt });
    return result.text;
  },
});

Key Principles

• Thread isolation: Each user/conversation gets its own thread for independent history
• Automatic message storage: Generated responses are automatically saved to the thread
• Context by default: Each generation includes recent message history automatically
• Async-friendly: Use actions for LLM calls; mutations for transactional writes
• Type safety: Always provide explicit return types to avoid circular reference errors

Common Patterns

• Per-user organization: Always include userId when creating threads
• Message history: Automatically included in LLM context
• Thread reuse: Same thread can be used by multiple agents

Next Steps

• Manage threads: See threads skill for conversation management
• Work with messages: See messages skill for saving and retrieving
• Add tools: See tools skill to let agents take actions