为什么值得关注

它提供了一个开箱即用的多平台部署方案，具备持久化向量记忆、沙盒化 Shell 执行、浏览器自动化以及统一的 LLM 供应商网关，全部通过容器化实现一致性。

适合谁使用

希望拥有高级功能 Telegram 机器人的开发者和 AI 爱好者
熟悉 Docker 和命令行的用户
希望通过统一接口尝试多个 LLM 提供商的人
倾向本地部署 AI 代理的隐私敏感用户

典型使用场景

通过浏览器控制自动执行网页抓取或会议管理等任务
在沙盒环境中管理文件并运行 Shell 命令
利用持久化记忆实现个人助手功能（如回忆过往对话）
跨不同 AI 提供商的模型对话与推理，支持回退

项目优势

多平台 Docker 部署（macOS、Windows、Ubuntu），并提供清晰的平台专属说明
持久化向量记忆（LanceDB），自动捕获和召回上下文
沙盒化 Shell 执行，确保文件与命令操作的安全性
通过单一 9Router 网关集成多个 LLM 提供商（NVIDIA、Groq、DeepSeek 等）

使用前须知

需要在宿主机上手动安装和配置 9Router
Linux 用户需额外步骤（--add-host）才能使容器连接到宿主机网络
未针对多用户访问进行加固，设计为单人操作的个人代理

README 快速开始

🦾 Mavent-claw

OpenClaw Agent Powered by 9Router — Multi-Platform Deployment Guide

📋 About

Mavent-claw is a personal deployment configuration for OpenClaw — an open-source, tool-enabled AI agent — connected through 9Router as a unified OpenAI-compatible gateway to multiple LLM providers (NVIDIA NIM, Groq, DeepSeek, MiniMax, and more).

The agent runs as a 24/7 Telegram-reachable automation agent, with browser control, filesystem access, persistent vector memory, and sandboxed shell execution — deployed via Docker for consistency across macOS, Windows, and Ubuntu.

✨ What's Configured

Feature	Provider	Status
💬 Chat / Reasoning	Custom combo via 9Router (multi-model fallback)	✅ Active
👁️ Vision (Image Input)	NVIDIA Nemotron 3 Omni via 9Router	✅ Active
🔍 Web Search	DuckDuckGo (experimental)	✅ Active
🧠 Persistent Memory	Memory LanceDB (vector-based, auto-capture + auto-recall)	✅ Active
🧮 Embeddings	`nemotron-3-embed-1b` via 9Router	✅ Active
🖥️ Sandboxed Shell	OpenShell Sandbox (mirror mode)	✅ Active
🌐 Browser Automation	Chrome-based control (incl. Google Meet integration)	✅ Active
📞 Voice Call	Mock provider (dev/testing only)	⏸️ Inactive (mock)
📱 Telegram	Paired via onboarding wizard	✅ Active

🛠️ Tech Stack

📦 Architecture

graph TB
    User[Telegram User] -->|Message| Gateway[OpenClaw GatewayDocker Container :18789]
    Gateway -->|OpenAI-compatible API| Router[9Routerlocalhost:20128]
    Router -->|Route| M1[NVIDIA NIM Models]
    Router -->|Route| M2[Groq Models]
    Router -->|Route| M3[DeepSeek / MiniMax]
    Router -->|Route| M4[NVIDIA Nemotron 3 OmniVision + Embeddings]

    Gateway --> Memory[(Memory LanceDBVector Store)]
    Gateway --> Shell[OpenShell Sandboxmirror mode]
    Gateway --> Browser[Chrome AutomationWeb + Google Meet]
    Gateway -->|Volume Mount| Workspace[(Local Workspace)]

    style Gateway fill:#f97316,color:#fff
    style Router fill:#6366f1,color:#fff
    style Memory fill:#10b981,color:#fff

Key components:

OpenClaw — the agent runtime, running inside a Docker container, exposing a local gateway on port 18789
9Router — runs natively on the host machine, a

项目描述

Personal deployment config for OpenClaw + 9Router — a 24/7 Telegram-reachable AI agent with browser automation, vector memory, and sandboxed execution, running via Docker on macOS, Windows, and Ubuntu.

Mavent-claw