mambo-agents 0.2.0a1

AI Agent framework on LangChain/LangGraph — multi-backend filesystem, sub-agent scheduling, conversation summarization, security review, and skills system out of the box.

Mambo Agents

AI Agent framework built on LangChain/LangGraph — multi-backend filesystem, sub-agent parallel scheduling, conversation summarization, security review, skills system, and more, out of the box.
基于 LangChain/LangGraph 的 AI Agent 框架 — 提供多后端文件系统、子代理并行调度、对话摘要、安全审查、技能系统等开箱即用的能力。

📖 English | 中文

This project draws architectural inspiration from deepagents, with independent refactoring and extensions. See comparison with deepagents.

Key Features

• Multi-backend Filesystem — StateBackend (in-memory), LocalBackend (local disk), SshBackend (remote SSH), HybridWorkspaceBackend (hybrid routing), unified through BackendProtocol
• Sub-agent System — sync/async sub-agents with parallel scheduling, streaming events, and isolated context windows
• Conversation Summarization — automatic long-history compaction with chained summaries and optional backend persistence
• Task Planning — MamboPlanMiddleware provides structured TODO lists, deeply integrated with the summarization system
• AI Security Review — pre-approve tool calls with a cheap model, escalating only high-risk operations to human review
• Skills System — progressive disclosure of skills, with multi-source overlay support
• Memory System — persistent context loaded from AGENTS.md, with AI self-learning write-back

Quick Start

pip install mambo-agents

from mambo_agents import create_mambo_agent, StateBackend
from langchain_core.messages import HumanMessage

agent = create_mambo_agent(
    "gpt-4o",
    backend=StateBackend(),
    include_general_purpose=True,
)
result = agent.invoke({"messages": [HumanMessage("Create a hello.py file")]})

Architecture

┌──────────────────────────────────────────────────────────┐
│                    create_mambo_agent()                   │
│                                                          │
│  ┌─────────────┐  ┌──────────┐  ┌────────────────────┐  │
│  │   Backend   │  │  Model   │  │    Middleware Stack │  │
│  │  Protocol   │  │ (LLM)    │  │                    │  │
│  │             │  │          │  │ BackendTools       │  │
│  │ State       │  │          │  │ Skills             │  │
│  │ Local       │  │          │  │ Memory             │  │
│  │ SSH         │  │          │  │ Summarization      │  │
│  │ TempWs      │  │          │  │ Planning           │  │
│  └─────────────┘  └──────────┘  │ SubAgents          │  │
│                                 │ AsyncSubAgents     │  │
│                                 │ SecurityReview     │  │
│                                 │ Patch + Reorder    │  │
│                                 └────────────────────┘  │
│                                                          │
│  ┌─────────────────────────────────────────────────────┐ │
│  │              LangGraph CompiledGraph                │ │
│  │  invoke() · astream() · astream_events()            │ │
│  └─────────────────────────────────────────────────────┘ │
└──────────────────────────────────────────────────────────┘

Docs

• Detailed Usage Guide — API reference, configuration, advanced usage
• Comparison with deepagents — architectural differences, feature mapping, design trade-offs

Related Projects

• MamboChat — Full-featured Web UI built on Mambo Agents

License

MIT License