zyf2007/ChatAPI

What it solves

ChatAPI allows humans to be called as if they were AI models. It provides an OpenAI-style API interface that routes requests from AI clients or agents to a human operator, enabling people to manually respond to prompts, act as a "model" within an agentic workflow, or serve as a Mock LLM during development.

How it works

The project implements a backend (Go) and a web console (React) that mimics the standard AI API endpoints. When an AI client sends a request to /v1/chat/completions or /v1/responses, the system routes the request to the human operator. The operator can then use the web console to respond to the user, assemble Tool Calling requests, or set up automatic reply rules.

Who it’s for

• AI Developers: Those building agents who need a Mock LLM to test workflows without spending tokens or using a real model.
• Agent Users: People who want to be integrated into an AI agent's logic as a human-in-the-loop component.
• Testers: Anyone wanting to simulate AI responses for specific scenarios.

Highlights

• API Compatibility: Supports OpenAI-style responses, Anthropic Messages compatible interfaces (/messages), and a custom /v1/responses endpoint.
• Web Control Console: Includes a React-based UI for managing responses and configuring automatic reply rules.
• Flexible Deployment: Supports Docker Compose one-click deployment, single binary builds with embedded frontend, or separate frontend/backend deployment.
• Database Support: Uses SQLite by default with planned support for PostgreSQL.
• Tooling Support: Ability to assemble Tool Calling requests via the console.