Saki AI Agent

<div align="center">

# Saki AI Agent
### A Local AI Agent Framework / Desktop-level AI Assistant / An Alternative to OpenClaw

</div>

<div align="center">

[![Node.js](https://img.shields.io/badge/Node.js-v18.0%2B-339933?logo=nodedotjs&logoColor=white)](https://nodejs.org/)
[![React](https://img.shields.io/badge/React-18.2-61DAFB?logo=react&logoColor=black)](https://reactjs.org/)
[![Vite](https://img.shields.io/badge/Vite-5.0-646CFF?logo=vite&logoColor=white)](https://vitejs.dev/)
[![Tailwind CSS](https://img.shields.io/badge/Tailwind-3.4-38B2AC?logo=tailwindcss&logoColor=white)](https://tailwindcss.com/)
[![Express](https://img.shields.io/badge/Express-4.21-000000?logo=express&logoColor=white)](https://expressjs.com/)
[![License](https://img.shields.io/badge/License-Apache%202.0-blue.svg)](LICENSE)
[![PRs Welcome](https://img.shields.io/badge/PRs-welcome-brightgreen.svg)](CONTRIBUTING.md)

</div>

Saki AI Agent is a **Local AI Agent, Desktop Application, and Windows AI Assistant** designed for Chinese users, emphasizing ease of use. It focuses on **Low-barrier Deployment, Low Token Cost, Low Resource Usage, and Enhanced Security by Default**: The application core can be deployed with just `2 hCPU / 200 MB`, delivers good results with small `8B-12B` local models, recommends the local model `glm-4.7-flash`, and provides **Sandboxed Execution, Sensitive Action Approval, and File Versioning**.

> **Keywords**: Local AI Agent, OpenClaw Alternative, Ollama, Windows AI Assistant, Desktop AI Workspace, Low-resource Deployment, Sandbox, Security, File Versioning



https://github.com/user-attachments/assets/91ac7121-2a01-4751-a0c3-bc257e097b0e


---

## 📖 Project Vision: A Local AI Agent Addressing Friction in Personal Desktop Scenarios
The year 2026 is the era of Agent proliferation, yet practical adoption remains challenging for average users: Online platforms, while convenient, raise privacy concerns, have capability limitations, and incur ongoing subscription/Token costs; Excellent projects like OpenClaw are powerful but lean more towards being a **self-hosted gateway / Agent runtime**, better suited for developers and power users rather than the average person who just wants a usable AI assistant on their local computer.

<div style="color: #ff6f61; font-weight: bold; text-align: left; margin: 20px 0;">Saki AI Agent's mission is to lower these barriers.</div>

It's not just a chat window; it's an **AI Copilot running on your local machine**: capable of web searching, reading local documents, invoking local image generation, connecting voice, and external channels. We aim to package cutting-edge AI capabilities into a **warm, user-friendly, cost-controllable, and more secure-by-default** desktop application.

---

## 🆚 Saki AI Agent vs. OpenClaw: Platform Layer vs. Product Layer

OpenClaw is a strong **self-hosted gateway / Agent platform**, with core strengths in multi-channel integration, plugin support, multi-Agent routing, and remote message ingestion.  
**Saki AI Agent** is more like a **Desktop AI Workspace** focused on the Chinese user experience and local usability.

The two are not opposites. OpenClaw leans towards a "platform foundation," while Saki leans towards the "end-user product layer." If your goal is:

*   To deploy an AI assistant on your own computer with a low barrier to entry.
*   To prioritize using local models, reducing reliance on cloud tokens.
*   To have a graphical configuration interface, file workflows, and an integrated experience for PPT/research/memory/third-party chat.
*   To have security policies that are more conservative and easier to understand by default.

Then Saki is often a more suitable choice than OpenClaw.

| Aspect | Saki AI Agent | OpenClaw |
| :--- | :--- | :--- |
| **Product Positioning** | Local Desktop AI Workspace | Multi-channel Self-hosted Gateway / Agent Platform |
| **Primary Focus** | Local Experience, Chinese Interaction, File Workflows, Desktop Usability | Channel Routing, Remote Access, Plugin Ecosystem, Multi-Agent |
| **Deployment** | Windows users can simply double-click `start.bat` / `configure.bat` | More CLI / Gateway / Channel / Workspace oriented; Windows usage often requires WSL2 |
| **Model Strategy** | Emphasizes local viability; good results achievable with small `8B-12B` models | Officially leans towards recommending the strongest latest-generation models, which in practice can lead to greater reliance on powerful cloud models |
| **Cost Control** | Suitable for long-term local residency; aims to minimize token consumption | Platform is powerful, but high-quality usage scenarios can more easily lead to ongoing token costs |
| **Security Experience** | Default permission modes, sandboxed execution, sensitive action approval, file versioning | Comprehensive security capabilities, but relies more on the operator to correctly configure trust boundaries, allowlists, tool policies, and sandboxes |
| **Target Users** | General Users, Indepe

... (truncated)