virgo777/buddyme

📚 What is buddyMe?

buddyMe is a Python‑based framework for building multi‑model AI agents that can automatically break down a complex request into smaller subtasks, run each subtask with a large language model (LLM) and a set of built‑in tools, then stitch the results back together. It ships with:

• Runtime hot‑switching among six Chinese LLM providers (GLM, DeepSeek, ERNIE, Qwen, Xiaomi, etc.)
• A skill system – 25+ pre‑written “skills” (API design, Python testing, markdown conversion, weather query, etc.) that can be loaded at three priority levels and reloaded without restarting.
• Tool integration – 8 ready‑to‑use utilities such as a bash executor, file read/write, grep, glob, Baidu search, and a generic invoke_skill wrapper.
• Persistent memory – user profile, conversation summaries and logs that survive across sessions, with decay/merge logic.
• Scheduling – /loop for periodic tasks and /cron for daily‑time‑specific jobs, each keeping a short execution history.
• Session statistics – /stats shows token usage, tool‑call distribution, skill usage and generated files, all computed locally (zero token cost).

The project also provides a desktop client (Windows installer) that bundles all dependencies and gives users a free daily quota of 45 million tokens / 1 000 calls.

---

🚀 Core Features (in plain language)

Feature	What it means for you
Multi‑model support	Switch between different LLM APIs on the fly; no need to restart the agent.
Three‑stage task flow	The agent first plans the work, then runs each sub‑task (LLM + tool), finally merges the results.
Tool system	Built‑in commands let the agent interact with the OS, read/write files, search the web, etc.
Skill system	Ready‑made knowledge modules (e.g., “write a Flask API”, “generate a PPT”) that the planner can call automatically.
Persistent memory	The agent remembers past interactions and can summarise or retrieve that memory later.
Scheduling (/loop / /cron)	Automate repetitive checks or daily summaries without manual prompting.
Zero‑cost stats	/stats reports token consumption and tool usage without spending LLM tokens.
Dual‑protocol adapters	Works with both OpenAI‑compatible and Anthropic‑compatible APIs, hiding the differences.

---

🛠️ Getting Started

1. Desktop client (recommended) – Download the Windows installer, run it, and you’re ready with a free token quota.
2. From source – Clone the repo, pip install -e ., create a .env file with the API keys for the models you intend to use, and run buddyme (or python -m buddyMe).
3. Interact – Type /help for the command list, then ask natural‑language questions or issue /model --switch deepseek to change the backend.

---

📂 Project Layout (high‑level)

buddyMe/
├─ agent_moudle/          # Core agent logic (planning, execution, merging)
├─ anthropic_standard/   # API wrappers for OpenAI‑ and Anthropic‑style endpoints
├─ cmd_library/           # Slash‑command parser (local, no token cost)
├─ initspace/             # Prompt templates (SOUL, IDENTITY, AGENT) and memory storage
├─ llm_moudle/            # Model configuration & hot‑swap handling
├─ skill_library/         # 25+ skill definitions
├─ tool_moudle/           # Implementations of bash, file I/O, search, etc.
└─ utils/                 # Helper utilities

---

📜 License

MIT – free to use, modify, and distribute.

---

In short: buddyMe is a full‑featured, locally‑run AI agent platform that lets developers compose LLM‑driven workflows, switch models on demand, and automate tasks with built‑in tools and skills—all while keeping usage transparent and cost‑free on the client side.