Local AI gets messy fast. One app runs the model, another handles chat, a terminal keeps the server alive, notes live somewhere else, and sooner or later you are copying context between all of them.
LM Nexus grew out of that problem: bring the pieces together without hiding the runtime underneath.
Simple should not mean opaque
Choosing a model and clicking Run should be enough when that is all you need. But when something breaks—or when exact runtime settings matter—you should also be able to see logs, endpoints, and launch options.
LM Nexus keeps both paths open: easy defaults when you want them, raw runtime control when you need it, and one workspace for the rest of the job.
Direct runtime/server access matters
LM Nexus can start a runtime for you without forcing inference through an extra Nexus proxy.
If you connect directly to a llama.cpp server started by LM Nexus, you are talking to llama.cpp itself. Nexus manages the process, but it is not sitting in the middle of the inference request.
That is useful for debugging, compatibility, and any workflow where direct access to the endpoint matters.
curl http://127.0.0.1:PORT/v1/chat/completionsLocal-first does not mean local-only
Local-first is the default, not a lock-in rule. LM Nexus can use local models, self-hosted runtimes, custom OpenAI-compatible endpoints, and connected API providers.
Some work belongs fully local. Some setups already have self-hosted inference. Other tasks need a connected provider. LM Nexus keeps those choices visible instead of hiding them behind one provider abstraction.
The model is only half the workflow
Running the model is only half the problem. Real work also needs files, notes, terminals, logs, agents, and reusable project knowledge.
LM Nexus keeps those things beside the model instead of cramming them into chat history. Agents can work with project files, notes can become reusable knowledge, and modules can add new tools without turning the app into one giant feature.
Current status
LM Nexus is still in private alpha, so there is no public download yet.
The site and docs are public early so you can see what is being built before the first release. Public builds will come after the private alpha is stable enough to support them.