WorkGPT is an agent framework for orchestrating LLM-driven task execution via API calls. It targets developers building autonomous agents that can interact with external services, offering a flexible way to integrate any API described by an OpenAPI specification. The primary benefit is enabling LLMs to perform complex, multi-step actions by abstracting API interactions into callable functions.

How It Works

WorkGPT leverages an agent-runner pattern. An OpenAiAgent (or similar LLM interface) receives a directive and a list of available APIs. APIs are exposed as local functions to the LLM, with authentication handled automatically based on provided keys. The LLM converses with itself, selecting and invoking appropriate API functions to fulfill the directive. A WorkGptRunner manages this interaction loop, and custom "finishing program" APIs can be defined to structure output.

Quick Start & Requirements

• Install: npm install workgpt
• Prerequisites: Node.js, an OpenAI API key (e.g., IPINFO_API_KEY).
• Example Usage: https://github.com/team-openpm/workgpt/blob/main/examples/workgpt.ts

Highlighted Details

• Supports any API with an OpenAPI definition, including those from the OpenPM package manager.
• Includes a TextBrowser API using Puppeteer for web crawling, extracting data from raw HTML text.
• Allows custom "finishing program" APIs with defined schemas for structured data extraction.
• Demonstrates agent interaction for tasks like IP address lookup and web scraping.

Maintenance & Community

• Project maintained by team-openpm.
• No explicit community links (Discord/Slack) or roadmap mentioned in the README.

Licensing & Compatibility

• License: MIT
• Compatible with commercial use and closed-source applications.

Limitations & Caveats

The framework relies heavily on the LLM's ability to correctly interpret OpenAPI specifications and generate valid API calls. The web crawling example uses Puppeteer, which may require additional setup and has its own resource footprint.