Convex Agents Tools
About
Convex Agents Tools enables agents to call external functions, APIs, and database operations through defined tools. Use it when your agents need to fetch data, perform actions, or integrate with external services, allowing for clean separation and multi-step operations. It supports creating human-in-the-loop workflows and autonomous agent decision-making.
Quick Install
Claude Code
Recommended/plugin add https://github.com/Sstobo/convex-skillsgit clone https://github.com/Sstobo/convex-skills.git ~/.claude/skills/Convex Agents ToolsCopy and paste this command in Claude Code to install this skill
Documentation
Purpose
Equips agents with the ability to take actions beyond text generation. Tools allow agents to call Convex functions, external APIs, and complex operations.
When to Use This Skill
- Agents need to query or modify database data
- Integrating with external APIs
- Creating human-in-the-loop workflows
- Agents autonomously deciding what actions to take
- Chaining tool calls for multi-step operations
Define Tools
Create Convex-aware tools:
import { createTool } from "@convex-dev/agent";
import { z } from "zod";
export const getUserDataTool = createTool({
description: "Fetch user information by email",
args: z.object({
email: z.string().email().describe("The user's email address"),
}),
handler: async (ctx, { email }): Promise<string> => {
const user = await ctx.runQuery(api.users.getUserByEmail, { email });
return user ? JSON.stringify(user) : "User not found";
},
});
Configure Agent with Tools
const agentWithTools = new Agent(components.agent, {
name: "Database Agent",
languageModel: openai.chat("gpt-4o-mini"),
tools: {
getUserData: getUserDataTool,
},
maxSteps: 5, // Allow tool calls
});
Enable Automatic Tool Calling
export const autonomousAgent = action({
args: { threadId: v.string(), request: v.string() },
handler: async (ctx, { threadId, request }) => {
const { thread } = await agentWithTools.continueThread(ctx, { threadId });
const result = await thread.generateText(
{ prompt: request },
{ maxSteps: 10 } // Allow up to 10 tool calls
);
return result.text;
},
});
Key Principles
- Use Zod for validation:
.describe()on fields helps LLMs understand parameters - Explicit return types: Always annotate handler return types
- Automatic history: Tool calls and results saved automatically in thread
- Context binding: Create tools inside actions where you have access to userId, etc.
Next Steps
- See fundamentals for agent setup
- See workflows for orchestrating multi-step operations
- See context for tool-aware context management
GitHub Repository
Related Skills
sglang
MetaSGLang is a high-performance LLM serving framework that specializes in fast, structured generation for JSON, regex, and agentic workflows using its RadixAttention prefix caching. It delivers significantly faster inference, especially for tasks with repeated prefixes, making it ideal for complex, structured outputs and multi-turn conversations. Choose SGLang over alternatives like vLLM when you need constrained decoding or are building applications with extensive prefix sharing.
evaluating-llms-harness
TestingThis Claude Skill runs the lm-evaluation-harness to benchmark LLMs across 60+ standardized academic tasks like MMLU and GSM8K. It's designed for developers to compare model quality, track training progress, or report academic results. The tool supports various backends including HuggingFace and vLLM models.
content-collections
MetaThis skill provides a production-tested setup for Content Collections, a TypeScript-first tool that transforms Markdown/MDX files into type-safe data collections with Zod validation. Use it when building blogs, documentation sites, or content-heavy Vite + React applications to ensure type safety and automatic content validation. It covers everything from Vite plugin configuration and MDX compilation to deployment optimization and schema validation.
llamaguard
OtherLlamaGuard is Meta's 7-8B parameter model for moderating LLM inputs and outputs across six safety categories like violence and hate speech. It offers 94-95% accuracy and can be deployed using vLLM, Hugging Face, or Amazon SageMaker. Use this skill to easily integrate content filtering and safety guardrails into your AI applications.
