Text, images and documents as first-class input. Tool calling, long-term memory, and a knowledge base — all running on your own machine. Open source, MIT licensed.
From a fresh download to a working multimodal agent.
One package, runs on your own machine.
Bring any model API, or run a local one.
Files, browser, shell, and your own skills.
Web UI, Telegram, WhatsApp and more.
Every capability below is already implemented and tested.
DeepSeek, Kimi, Qwen, OpenAI, NVIDIA NIM and local Llama models.
LiveSearch, code execution, sandboxed commands, file read/write, pipelines, sub-agents.
LiveCommand whitelist + Docker isolation + path constraints.
LiveSupports reasoning models such as DeepSeek R1 and OpenAI o1.
LiveSee images, read documents, and understand screenshots as a first-class input.
LiveLocal document indexing plus vectorised retrieval.
LiveThe phone runs no model. Sign in with the same account and you are controlling the very same agent on your own machine or server.
LiveThe phone itself produces zero compute. It is a projection of your account, not a second agent.
One memory, one set of agents, one workspace and one file tree. Not a copy — the same thing.
Model inference and every tool call run on your PC or your server. The phone only renders and sends commands.
A job started on your desktop keeps running on the engine. Watch its progress from the phone and keep steering it.
Chat, agents, workflows, files, tools, multi-agent collaboration — exactly what you get when self-hosted.
Same account required. If you sign in with a different one, you are of course talking to a different agent.
Every line below is implemented and tested.
| Capability | Business rule | Status |
|---|---|---|
| Multimodal input | Text, images and documents as first-class input | Live |
| Tool calling | Files, browser, shell and custom skills | Live |
| Long-term memory | Remembers you across sessions and restarts | Live |
| Knowledge base | Local document indexing with vectorised retrieval | Live |
| Model switching | Any OpenAI-compatible API, plus local models | Live |
| Multi-channel | Telegram, WhatsApp, Discord, Slack and more | Live |
| Voice | Speech-to-text and text-to-speech | Live |
| Self-hosting | Run it on your own machine or your own server | Live |
This is a platform for users worldwide.
Your data never leaves your machine. Nothing is uploaded to us.
You own the deployment: the model keys, the database and the logs are all under your control.
Open-source and self-hosted: your data stays on your own machine.
Mainland restrictions on cross-border settlement and data export are severe.
Please note:
Open-source nature is unchanged:
Engine evaluation: The engine runtime is a complete implementation, not a prototype. Multimodal input, tool calling, long-term memory and knowledge-base retrieval are all implemented and tested, and everything runs inside a single installation.
Open source, free forever.
Straight answers.
Install with a single PowerShell / Shell command.
irm https://lingxi-engine.com/install.ps1 | iex
curl -fsSL https://lingxi-engine.com/install.sh | bash
git clone https://github.com/lingxi-engine/lingxi-engine.git && cd lingxi-engine && npm install
These commands point at the official site (Cloudflare-protected). If they fail, your network is likely blocking the script host — use the manual archive on the download page.
Questions, skill requests, agent ideas.
Technical questions, feature requests, bug reports.
Setup guides, model configuration, tool integration and channel setup.
Read CONTRIBUTING.md. Issues and pull requests are both welcome.
This is an open-source project built by one person. Open source means you can self-host, self-modify, and self-decide.