Your local models, agents and GPUs. One Windows control center.
Русский · Українська · English
Download preview · Getting started · Report an issue
3.0.0-alpha.1 · Early preview · Windows 10/11 x64 · MIT
Station brings model launch, agent interfaces, hardware status and tray controls into one desktop application. It works with CPU-only systems and different GPU counts; GPU features depend on the hardware and driver capabilities.
| Area | What you can do |
|---|---|
| Dashboard | See model/server/agent status and GPU telemetry; launch and stop components |
| Agents | Discover Qwen Code/Desktop, Hermes, Pi Coding Agent, OpenClaw and Aider; open official installers/releases; manage owned processes |
| Models | Configure llama.cpp/llama-swap profiles, select devices, launch and unload local models |
| Tray | Use quick controls and two configurable metric channels with numeric values and background fill |
| Services | Manage local processes; check remote ComfyUI/HTTP workers with optional monitoring |
| Startup | Choose Windows startup separately from services, model and agent startup; set a delay and cancel |
Download LocalAgentAIStation-3.0.0-alpha.1-Setup-x64.exe from Releases. Choose English, Russian or Ukrainian in Setup. Python is bundled; no administrator rights are required. The application UI is currently Russian. Agents, model weights and inference engines are separate installations. First launch does not automatically start them.
The program uses %ProgramFiles%\Local Agent AI Station; its data uses
%LOCALAPPDATA%\LocalAgentAIStation. Start-menu and optional desktop shortcuts,
Windows uninstall and in-place updates are supported. See the
quick start for setup, updates and removal.
- This release is unsigned. There is no automatic updater yet.
- GPU monitoring is included. TCC/WDDM changes require compatible NVIDIA cards and an optional privileged helper, which Setup does not install. Installed-service end-to-end acceptance is still pending; see GPU helper.
- Physical testing used CPU, one selected NVIDIA GPU and two NVIDIA GPUs on one PC. Tests simulate 0/1/2/3/4/8 GPU topologies; other physical configurations are not certified.
- Qwen Desktop, Pi and OpenClaw have integration evidence; adapter coverage does not guarantee compatibility with every upstream release. See validation.
- Agent image-generation tool integration and prolonged maximum-context qualification remain work in progress. A configured remote service is not automatically an agent tool.
See CONTRIBUTING, architecture, release procedure and roadmap. Development checkpoints live in HANDOFF; private machine notes, settings and test evidence stay outside Git. Report bugs in English, Russian or Ukrainian.
python -m venv .venv
.\.venv\Scripts\python.exe -m pip install -r requirements-dev.txt
.\build_installer.ps1 -Python .\.venv\Scripts\python.exe -ISCC 'C:\Program Files (x86)\Inno Setup 6\ISCC.exe'Commit your source changes before packaging. Build dependencies and setup details
are described in RELEASING. The source can also be run with
.venv\Scripts\pythonw.exe main.pyw --no-startup.
MIT license. Bundled dependencies retain their own licenses; see third-party notices. Station is an independent project and is not affiliated with the agent or model vendors.