virgo777/buddyme

What it solves

buddyMe is a multi-model AI agent framework designed to handle complex tasks by automatically breaking them down into smaller, manageable sub-tasks. It solves the problem of model lock-in by allowing developers to switch between different LLM providers at runtime without interrupting the session, while providing a structured system for skills, tools, and long-term memory.

How it works

The framework operates through a three-stage execution process:

1. Planning: The agent analyzes the request and uses skill metadata to decompose the complex task into a sequence of sub-tasks.
2. Execution: Each sub-task is executed independently using a loop of LLM calls and tool usage, with specific skills injected as instructions.
3. Merging: The results of all sub-tasks are combined into a final output.

It utilizes a layered system prompt (Personality, Identity, Behavior, and Tool Schemas) and a command system (using / prefixes) that handles administrative tasks locally to save tokens.

Who it’s for

It is built for developers who need a flexible, autonomous agent capable of acting as a programming assistant or a general-purpose task proxy, especially those who want to integrate multiple LLM providers and automate recurring tasks.

Highlights

• Multi-Model Hot-Switching: Supports seamless runtime switching between providers like GLM, DeepSeek, ERNIE, Qwen, and MiMo.
• Tool & Skill System: Includes 8 built-in tools (bash, file editing, search, etc.) and over 25 pre-set skills for domains like API design and Python testing.
• Persistent Memory: Maintains user profiles, conversation summaries, and logs across sessions with support for memory decay and merging.
• Scheduled Tasks: Features both interval-based loops (/loop) and daily fixed-time triggers (/cron) with execution history tracking.
• Token Efficiency: Local command processing and session statistics (/stats) ensure transparency and reduced API costs.