fumingyang2004/Tulpa

What it solves

Tulpa acts as a bridge between your local messaging apps (QQ and WeChat) and AI agents. It solves the problem of data silos by allowing you to securely index, search, and retrieve chat history, files, and images from these platforms, while also enabling AI agents to participate in QQ group chats with custom personas.

How it works

Tulpa runs locally on your machine and interfaces with your messaging clients to ingest data. It exposes this data to external AI agents via the Model Context Protocol (MCP). You can define specific scopes—such as which chats, dates, or file types an agent can access—and grant granular permissions for actions like reading, sending messages, or managing groups. For real-time group interaction, it integrates with OneBot services to handle live events.

Who it’s for

This project is for power users and developers who want to integrate LLMs into their daily communication workflows. It is ideal for those who need to summarize long chat histories, retrieve specific files from group conversations, or deploy AI-driven personas to interact in QQ groups.

Highlights

• Privacy-focused: Data is processed and stored locally; you control exactly what information is shared with external AI models.
• MCP Support: Uses the Model Context Protocol to connect with popular AI clients like Claude Code, DeepSeek, and others.
• Custom Personas: Supports custom character prompts for group chats, allowing agents to adopt specific personalities and learn from group context.
• Granular Control: Provides fine-grained authorization for read/write actions, ensuring agents only perform tasks you have explicitly approved.