Zact — The Standard Library for LLM Actions

Zact equips your LLM agents with real-world capabilities via ready-to-use tools — with zero boilerplate.

What Is Zact?

Zact is an NPM package offering a library of predefined tools (or "actions") that can be invoked by LLMs such as OpenAI's GPT, Anthropic Claude, or local models via function-calling or tool-use APIs.

Each tool includes:

• A real function (run) to execute the action (e.g., get weather, send email)
• A JSON schema for structured input
• Compatibility with OpenAI and other function-calling interfaces

Installation

npm install zact

Quick Example (OpenAI GPT-4 Function Calling)

import { tools, providers } from "zact";

const response = await providers.openai.callLLMWithTools({
  prompt: "What's the weather like in Nairobi, Kenya?",
  tools: [tools.getWeather],
  model: "gpt-4-0613",
  apiKey: process.env.OPENAI_API_KEY
});

console.log(response.result);

Zact handles the function call logic — the model simply chooses the right tool and parameters.

Built-in Tools

| Tool | Description | | ------------- | -------------------------------------- | | getWeather | Get current temperature for a location | | emailSender | (Coming soon) Send email via API/SMTP | | webSearcher | (Coming soon) Perform a web search | | clipboard | (Planned) Copy text to clipboard | | ... | Additional tools on the way |

Tool Structure

Each tool is defined using registerTool() and includes:

• name — the tool identifier
• description — visible to the LLM
• parameters — a JSON schema defining input
• run(args) — the logic to execute

Example (tools/getWeather.ts):

export const getWeather = registerTool({
  name: "getWeather",
  description: "Get the current temperature for a given location.",
  parameters: {
    type: "object",
    properties: {
      location: {
        type: "string",
        description: "City and country, e.g., Paris, France"
      }
    },
    required: ["location"],
    additionalProperties: false
  },
  run: async ({ location }) => {
    // Fetch lat/lon and call weather API
    return `The temperature in ${location} is 27°C`;
  }
});

Supported Providers

Currently supported:

• OpenAI (callLLMWithTools) — supports functions[] interface

Planned:

• Anthropic Claude (tool_use API)
• Local models (e.g., LLaMA) via prompt parsing

Add Your Own Tools

Define a new tool using registerTool:

export const sayHello = registerTool({
  name: "sayHello",
  description: "Say hello to someone",
  parameters: {
    type: "object",
    properties: {
      name: { type: "string" }
    },
    required: ["name"]
  },
  run: async ({ name }) => `Hello, ${name}!`
});

Then register and export it from index.ts.

Why Zact?

| Feature | Benefit | | ---------------------- | --------------------------------- | | Prebuilt tools | Skip repetitive schema + logic | | Minimal setup | One-liner integration | | Multi-model support | Compatible with various LLMs | | Extensible | Add and share your own tools |

Pitch Line

Zact: Plug-and-play tools for your AI agents.
Auto-generated schemas, real-world functions — minimal effort.

Contributing

Contributions are welcome!

1. Fork the repository
2. Add your tool in tools/
3. Export it in index.ts
4. Write a test (if possible)
5. Submit a PR

License

MIT — free to use, modify, and distribute.