@QingQ77: Describe requirements in natural language, and the AI Agent automatically breaks down steps, calls tools to complete development, file operations, browser control, and other tasks, while also providing a full-fledged editor and terminal. https://github.com/Liuchun-oss/codelf-agent… Codelf is…

Describe your needs in natural language, and the AI Agent automatically breaks down steps, calls tools to complete development, file operations, browser control, and other tasks, while providing a full-featured editor and terminal. https://github.com/Liuchun-oss/codelf-agent… Codelf is a desktop AI assistant built with Electron. You tell it what you want to do, and it breaks down tasks, reads files, runs commands, opens browsers, and operates desktop applications — all the way through. It’s also a proper editor with code completion, terminal, and Git. Supports models like DeepSeek, Claude, ChatGPT, etc., network-friendly in China, and specially optimized cache hit rate to save API costs. Supports plugins and MCP extensions. Works on Windows and macOS. — # Liuchun-oss/codelf-agent Source: https://github.com/Liuchun-oss/codelf-agent # Codelf (码灵) An AI assistant that handles anything with natural language Just tell it what you want to do, and let it handle the rest. Codelf is a desktop application with a built-in autonomous AI Agent. It can help you develop projects, organize materials, and operate your computer using plain language — what people call vibecoding. It’s also a fully functional editor: write code, run terminals, connect to browsers, and control local programs. version license platform website electron react typescript Preview · Why Choose · What It Can Do · Tech Stack · Quick Start · Build · Website (https://codelf.top) · Contact — ## Preview ## Why Choose Codelf In a nutshell: a natural language AI assistant that works directly in China, supports multiple major models, and is particularly cost-effective. You don’t need to know code or learn complex operations — just describe what you want in plain language, and it breaks down the steps, calls tools, and gets things done. Whether it’s developing a project, editing a document, running some data, or operating a program on your computer, leave it to Codelf. - Especially friendly to networks in China. Natively supports models like DeepSeek that are directly accessible in China — no need for VPNs to run AI. Also compatible with any OpenAI-compatible API: want to use a domestic relay, self-built gateway, or company intranet proxy? Just configure it and you’re good to go. - Native support for major models, switch anytime. DeepSeek, Claude, ChatGPT, Azure OpenAI are all built-in. Just fill in your API Key and start using. Use the cheapest one, the smartest one, switch whenever you want — no subscription lock-in. - High context cache hit rate, really saves money. We’ve specially optimized the conversation organization, generating stable cache keys for each session so that model providers’ prompt caching hits as often as possible. For the same long context, the cost difference with/without cache can be many times. Over long coding sessions, the savings are real. - Can control browsers and desktop applications. Not just writing code — it can open web pages, click buttons, fill forms, scrape content, and also launch desktop programs on your computer, read their UI, click, screenshot — like a true computer butler doing errands for you. - Lightweight, fast startup. No bloated plugin marketplace or background services. Out of the box, conversation, editing, terminal, and AI Agent are all ready. - Your data stays on your machine. Pure local desktop app; code and files never pass through any third-party hosting — safe and under your control. The knowledge base vector database is also fully local: no uploading enterprise documents, no internet connection — zero risk of sensitive data leakage. - A powerful assistant for enterprise knowledge management. Built-in local RAG (Retrieval Augmented Generation) system. Supports importing internal technical documents, business manuals, product specifications, historical project materials. When answering questions, the AI automatically retrieves relevant content and cites it, making the AI assistant truly understand your business. Tables, flowcharts, parameter sheets, and other structured information are fully preserved. Chinese semantic search is far more accurate than keyword search. New employee training, technical Q&A, specification lookups — efficiency is doubled. - Both an assistant and a handy tool bench. Monaco editor, LSP intelligence, integrated terminal, Git, global search, command palette — everything you need. When you want to take over, you can. - Open source and free. MIT license — you can freely read, modify, and distribute. ## What It Can Do - Get things done for you: Describe your needs in one sentence, and the autonomous AI Agent reads/writes files, runs commands, searches info, calls tools — carrying out multi-step tasks from start to finish, whether it’s developing a project or handling daily chores. - Plan ahead: Complex tasks first enter a read-only Plan mode to research and propose a solution. Only after confirmation does it start working. Each round has checkpoints; if something goes wrong, you can roll back entirely. - Enterprise knowledge base RAG: Built-in local vector database supports importing common enterprise document formats such as docx / xlsx / xls / pdf / doc / md / txt. Tables are fully preserved, automatically chunked and vectorized, so the AI can retrieve company standards, technical documentation, and business materials — becoming a smart assistant that truly knows your business. Especially suitable for: - 📋 Technical documentation management: API docs, interface specs, architecture design documents — AI can directly cite them to answer technical questions. - 📊 Business knowledge consolidation: Product manuals, operation workflows, data dictionaries — new hire training materials instantly become an AI assistant. - 🔍 Compliance and standards: Company policies, industry regulations, quality standards — automatic retrieval to avoid omissions. - 💡 Project experience repository: Historical project docs, problem solutions — avoid repeating mistakes. - Fully localized, data never leaves the corporate intranet. Supports Chinese semantic search (built-in bge-small-zh-v1.5 model) with recall accuracy far exceeding keyword search. - Control browsers: Built-in web search, web scraping, and Playwright browser automation — can open web pages, navigate, click, fill forms, take screenshots, and scrape content on its own. - Control desktop applications: Cross-platform control of local programs on your computer — launch/close apps, read window UI, click buttons, fill text, take screenshots — let it handle manual operations. - Extensible: Supports Agent Skills for one-click installation of reusable skills, and can integrate third-party tools and resources via MCP (Model Context Protocol). - Professional coding: Built-in language services for TypeScript / Python / Vue / JSON / YAML, etc. Inline completions, selected text rewrite, one-click error fix, auto-generate commit messages. Automatic detection and switching of Python environments. UI and editor are natively localized in Chinese. ## Tech Stack | Category | Technology | | — | — | | Desktop Framework | Electron 33 | | Build Tools | electron-vite, Vite 5 | | Frontend | React 18, Zustand, TypeScript 5 | | Editor | Monaco Editor, Shiki | | Terminal | xterm.js, @lydell/node-pty | | AI SDK | @anthropic-ai/sdk, openai, @modelcontextprotocol/sdk | | Knowledge Base RAG | better-sqlite3, sqlite-vec, @huggingface/transformers (local vectorization), mammoth (docx), xlsx (Excel), pdfjs-dist (PDF) | | Language Services | basedpyright, typescript-language-server, vscode-langservers-extracted, yaml-language-server, @vue/language-server | | Browser Automation | Playwright | | Packaging | electron-builder | ## Quick Start ### Prerequisites - Node.js >= 18 - npm (or pnpm / yarn) ### Install Dependencies bash npm install ### Development Mode bash npm run dev Starts the app window with hot reload. ### Configure AI Model On first use, go to “Settings” in the app and enter your API Key for the provider (e.g., DeepSeek, Anthropic, OpenAI, etc.). Then you can start using the AI Agent capabilities. ## Build and Package bash # Windows installer (NSIS) npm run dist # Windows portable directory npm run package # macOS DMG (supports arm64 / x64) npm run dist:mac Build outputs go to the release/ directory. ## Contact For questions, suggestions, or just to connect, feel free to scan and add me: | WeChat | QQ | | :—: | :—: | | | | ## License This project is open source under the MIT license. Copyright © 2026 巫枫