Neo.mjs
- The Application Engine for the AI Era
- Conversational UIs
- Agentic Control
- The Neural Link
- Live Runtime Mutation
- AI-Native Runtime
- JSON Blueprints since 2019
- Object Permanence
- Context Engineering
- True Multithreading
- SharedWorker Simulations
- Physics-Driven UI
- Desktop-Class Performance
- Linear Scalability
- Multi-Window Orchestration
- Cross-Window Messaging
- Hierarchical State Management
- Clean Architecture
- Separation of Concerns
- Zero Build Step
- 100% Web Standards
- No Transpilation
- Instant Page Reloads
Neo.mjs is the Application Engine for the AI Era. Build desktop-class, multi-window apps where AI agents can see, understand, and control the runtime—enabling the next generation of Conversational UIs.
Amazing Potential
Shared Socket Data
Click the start button to see the view reflect changes in the data. And the app is multi-window: the table and charts can be undocked into their own windows. In fact, the entire demo can be undocked. All windows share the same data - less traffic and costs.
Extreme Speed
40,000 Updates /s
This demo shows the Neo.mjs helix component, along with a "Helix Controls" panel. Move your cursor over the helix, then rapidly scroll left and right to rotate, and up and down to zoom. As you do, look at the delta updates counter at the top. Neo.mjs easily handles 40,000 updates per second, and beyond.
How?
How does Neo.mjs do it?
When a Neo.mjs app launches 3+ webworkers are spawned: one that holds App logic, one for calculating delta DOM updates, and one for backend calls. Each webworker runs in its own thread, and thus, in its own processor core. This means these processes run in parallel: DOM updates and transitions are not affected by your App logic and always run smooth. Every processor-intensive task runs outside the Main Thread.
Threading Subsystem
- True Multithreading: App, Data, VDOM, and Canvas live in separate workers.
- The Main Thread is treated as a "dumb" renderer, ensuring 60fps fluidity.
- Eliminates UI jank by isolating logic from rendering.
- Includes a dedicated Task Worker for background processing.
Rendering Pipeline
- Asymmetric VDOM: Diffing happens off-thread.
- Sends compressed JSON deltas to the Main Thread.
- Faster, more secure, and AI-friendly than main-thread diffing.
- Supports over 40,000 delta updates per second.
The Scene Graph
- Components are persistent objects, not ephemeral render results.
- Retain state and identity even when detached from the DOM.
- Enables Runtime Permutation: Drag active dashboards between windows.
- The "Lego Technic" model vs. "Melted Plastic" (other frameworks).
State Subsystem
- Built-in, hierarchical State Providers.
- Push and Pull reactivity models.
- Components bind declaratively to state changes.
- Updates are handled efficiently off the main thread.
Multi-Window Orchestration
- A single engine instance powers multiple browser windows.
- SharedWorkers enable real-time state synchronization.
- Move components between windows without losing state.
- Ideal for Trading Platforms and Control Rooms.
Zero-Build System
- Runs as native ES Modules directly in the browser.
- No transpilation or bundling required in Dev Mode.
- Instant reloads: What you write is what you debug.
- 100% aligned with modern Web Standards.
Code at the Speed of Thought
AI agents can now "see" and "touch" your running Neo.mjs app in real-time via the Neural Link.
The Neural Link
- Introspect the live Scene Graph.
- Live Runtime Mutation (Hot-Swapping).
- Simulate user interactions for E2E.
- Bi-directional agent communication.
MCP Infrastructure
- Knowledge Base: Semantic codebase search.
- Memory Core: Persistent agent context.
- GitHub Workflow: Autonomous PR reviews.
- Neural Link: The Bridge to the runtime.
Context Engineering
- Advanced Code Execution: Orchestrate tools as APIs.
- Tool Composition: Chain complex workflows locally.
- On-Demand Loading: Prevent context window overload.
- 181 tickets resolved in 15 days.