Product tour

One place for the model
and the work around it.

Run a model, give it tools, keep project files and notes nearby, or connect it to devices and persistent worlds. LM Nexus brings those pieces together without hiding how they work.

Open the docs Explore the domains
The Nexus 00

The model is only one piece.

Files, agents, devices, and worlds keep their own state. LM Nexus gives the model controlled ways to work with them instead of pushing everything into one chat or backend.

Models & Runtime Agents & Tools Workspace Physical Systems Living Worlds LM Nexus one workspace · modular underneath
LM Nexus one workspace · modular underneath
Five areas share one LM Nexus workspace, while each keeps control of its own data and behavior.
Current domains 01—06
01 Current

Models & Runtime

Run local models without giving up control.

Download GGUF models, save profiles, start llama.cpp, watch logs and metrics, or connect directly to the server when you want full control.

Explore this realm →
Models →Runtimes →Download a model →Use the local API →
  • local GGUF library and Hugging Face downloads
  • profiles, variants, groups and runtime settings
  • start/stop control for llama.cpp plus direct server access
  • local /v1 serving plus self-hosted and connected providers
02 Current

Agents & Tools

Let models do work, with limits you can see.

Choose the model, tools, skills, and MCP servers an agent can use. Agent Runtime handles tool calls, approvals, questions, and delegation while you stay in control.

Explore this realm →
Agents →Configure an agent →Run an agent safely →
  • saved persona and Agent profiles
  • tool allowlists and approval modes
  • skills, MCP assignments and default context
  • approval/question cards tied to exact Runtime interactions
03 Current

Workspace

Keep the work beside the model.

Files, Git, Terminal, Notes, and Knowledge live beside chat and agents, so you do not have to keep moving project context between separate apps.

Explore this realm →
Workspace →Work with files →Build Knowledge →
  • project-aware Files workspace and editor
  • local terminal, logs and Git surfaces
  • Markdown Notes for capture and working material
  • structured Knowledge guides for reusable reference
04 Current

Physical Systems

Connect models to real devices.

Device Nodes let LM Nexus discover local hardware, read its status, and expose selected actions to the UI or to agents.

The current device layer is hardware-agnostic. Broader robotics and protocol-specific support is still future work.

Explore this realm →
Browse current documentation →
  • private-LAN node admission and inventory
  • health, event and capability discovery
  • controlled device actions
  • module-owned widgets and Agent Tool exposure
05 Current

Living Worlds

A language model can participate in a world, not just a conversation.

Nexusverse keeps characters, relationships, timelines, lore, and other world state in a local database. The model can help narrate or propose changes, but those changes stay reviewable.

Autonomous background simulation and deeper simulation mechanics are still future work.

Explore this realm →
Browse current documentation →
  • Storyteller, Lorebook and relationship graph
  • Timeline, Assets and Atlas
  • checkpoints and recovery
  • reviewed World Turns, typed proposals and durable foreground batches
06 Current

Modular platform

One app, modular underneath.

LM Nexus can grow through modules without turning every feature into a separate app. Modules add pages, settings, tools, panels, and integrations through a shared extension system.

Module model →Marketplace & extensions →Manage extensions →
  • module manifests and product bundles
  • declared capabilities and dependencies
  • Marketplace installs, updates, rollback, and trust checks
  • restart/reload/build-aware activation instead of unsafe hot-loading
What is being built next 07
Preview

Cognitive runtime

Keep longer work alive between turns and restarts.

A newer runtime layer is being integrated for longer tasks. It adds persistent work state, memory, context building, model routing, recovery, human handoffs, and deeper delegation. These pieces work in Preview, but they are not yet the default runtime path.

  1. 01
    Durable Work Runtime Keep goals, tasks, outputs, and verification attached to longer-running work.
  2. 02
    Memory + Context Builder Remember useful information without dumping all of it into every model request.
  3. 03
    Model Router Choose a model based on what the task actually needs instead of always using one default.
  4. 04
    Runtime Supervisor Decide when to retry, recover, switch approach, or ask for help.
  5. 05
    Human Tasks + delegation Pause for a person when needed and keep delegated work attached to the parent task.
Read the Cognitive Runtime overview → See the roadmap →

Go deeper

Start with what you want to do.

The docs walk through everyday tasks such as running a model, connecting a provider, configuring an agent, working with files, and managing extensions.

Getting started Browse docs