An AI assistant for everyone — chat, code, tools and models, organised into one clear, quiet entry point that's always ready.
Read the docs →Turn this meeting recording into key points and action items.
Done. ① Set the Q3 growth target; ② marketing to ship the campaign this week; ③ 3 action items assigned by owner.
Product Demo
Why NearWork
Compatible with OpenAI, Anthropic & Gemini protocols. Connect any model on the market. 7 purpose-built model slots with automatic routing.
Code editing, file ops, Git, SSH, Office docs, PDF & OCR — covering the full dev workflow.
AI sees the screen and drives the mouse & keyboard to operate desktop apps for you.
Code gen & review, bug fixing, docs, data analysis, translation — ready to use.
Connect any external tool via MCP; up to 16 concurrent sub-agents.
Continuous voice conversations. Hands-free, just speak to get work done.
Assign work straight from Feishu / WeCom; Desktop · Web console · CLI · VS Code share one engine, with unattended scheduled tasks.
Encrypted API keys, 4-tier permissions, dangerous ops auto-blocked. Your files and keys never leave your machine.
A multi-agent team: a leader plans while each member carries its own role, skills and MCP servers — work is split, run in parallel and summarised.
Everywhere You Work
Windows / Mac / Linux, full experience
Dev-first, pipe & script integration
Inline completions, review, refactor
Remote engine, team & API access
NearWork Assistant · Windows / Linux / macOS desktop apps
Or install CLI via npm
$ npm install -g @lingxi-agent/core
$ lingxiRequires Node.js 22+
What people ask most before installing: which models work, whether you need your own API key, and where your data lives.
NearWork is a multi-model desktop AI agent, not just a chat window: it reads and writes local files, runs commands, fetches web pages and calls tools and skills, turning a one-line request into a finished result.
Over the OpenAI, Anthropic and Gemini API protocols, NearWork connects to 16 model providers, including DeepSeek, Qwen, Kimi, Zhipu, Doubao, MiniMax, ERNIE, Hunyuan, StepFun, OpenAI, Anthropic, Gemini, xAI, Groq and Mistral.
No. You can use the official model credits, or enter your own API key in settings, in which case requests go straight to the provider you chose.
Configuration, sessions, skills, the knowledge base and memory live in a local data directory (defaults to ~/.LingXi, overridable with the LINGXI_HOME environment variable). Only the inference requests sent to your chosen provider leave the machine, and API keys are encrypted locally.
Yes, the free plan is available on sign-up and includes official model credits; see the pricing page for more credits and team features.
Yes. Any MCP server can be added through mcp.json, with user-level and project-level configuration, plus a plugin marketplace.