Your files stay yours
Point the library at your existing model directories. Downloads stay plainly named, on your disk. Remove Eugene Plexus and your model files are still right where you put them.
Your library, without the lock-in.Local-first. Open source. In development.
Eugene Plexus.
Your models. Your hardware. One control plane.
Find and run language models. Manage your engines and route requests through one OpenAI-compatible endpoint, all from your browser.
No public platform release yet. Follow the work on GitHub.
01 / The platform
Your inference setup,
working together.
The models are here. The engines are capable. Eugene Plexus brings the operations together: discover, configure, supervise, and route, without giving up control of your setup.
A browser for the whole setup
Manage runtimes, profiles, and configuration from one interface. Built-in authentication supports access over your tailnet, including headless machines.
Local hardware. Network-aware control.Works with proven engines
Eugene Plexus supervises llama.cpp and vLLM instead of replacing them. An OpenAI-compatible gateway gives your tools a consistent way to reach the backends you choose.
The operations layer, not another engine.02 / The workflow
Your models.
Your engines.
One control plane.
From a model on disk to a service your tools can use.
A model library, supervised runtimes, and a routing gateway bring the pieces together. Keep the engine choices in your hands.
Explore the architectureFind your next model
Search the catalogue, read model cards, and choose a quantization with hardware-aware guidance. Download with progress and resume, or use models already on your disk.
Make the setup yours
Keep named launch profiles for each model. Configure your runtimes through a browser interface, with settings drawn from the components themselves.
Run engines you trust
Start, stop, and inspect llama.cpp and vLLM runtimes. Eugene Plexus supervises the processes, with idle unloading and wake-on-demand support.
One endpoint, more options
Connect your tools to one OpenAI-compatible endpoint. Route across local runtimes and optional cloud providers, with replica balancing and priority-based failover.
See what actually happened
Try a model in the playground. Inspect request metrics, latency, throughput, and which backend served a request, including the attempts that failed.
03 / Built in the open
Early days.
An open invitation.
Eugene Plexus is under active development. There isn't a public platform release to download yet, but the code and the work are out in the open.
Explore the repositories, follow development, or help shape what comes next.
Find us on GitHubProject status In development
Core licenseApache 2.0
Built forLocal LLM inference
DevelopmentPublic repositories
04 / A few details
Good questions.
Is this a chatbot or an inference engine?
Neither. Eugene Plexus is the control plane around your inference setup. It supervises upstream engines, manages model files and settings, and routes requests. The playground is one way to use it, not the whole product.
Can I download it yet?
There is no public platform release yet. Developers can explore the source and the current setup instructions in each repository. This is active development, not a finished installation experience.
Which inference engines does it support?
The current integrations supervise llama.cpp and user-installed vLLM. MLX is planned, not available yet. Hardware support depends on the engine and model; consult the project documentation for verified configurations and current limitations.
How can I get involved?
Browse the GitHub repositories, read a component's contributing guide, and open an issue with a bug report or an idea. Contributions use the Developer Certificate of Origin, with signed-off commits.