Screenshot del progetto
Panoramica
Neo.mjs is an open-source JavaScript framework structured as two complementary hemispheres: The Brain (an Agent OS running in Node.js) and The Body (a multi-threaded application engine running in the browser). The Brain orchestrates a cross-model AI swarm—drawing on models from Anthropic, Google, OpenAI, and Moonshot—to autonomously run the engineering lifecycle, including ideating, building, and reviewing code. The Body is an off-main-thread runtime utilizing Web Workers for application logic, VDOM diffing, data processing, and canvas rendering. A 'Neural Link' possession interface bridges the two, allowing AI agents to inhabit and mutate live applications in real time.
Funzionalità principali
- Cross-model AI swarm coordination (Claude, Gemini, GPT, Kimi)
- Active Hybrid GraphRAG and Native Edge Graph for persistent memory
- Neural Link possession interface for live runtime mutation
- DreamService REM-cycle consolidation for priority re-steering
- Off-Main-Thread architecture (App, VDom, Data, Canvas, Shared Workers)
- Zero runtime dependencies and native ES Modules without transpilation
- Multi-tenant cloud deployment for pointing the Agent OS at external codebases
- Self-healing loops that convert runtime failures into fixes and memory
Requisiti, installazione e avvio rapido
Utilizzo
Compatibilità del modello e casi d'uso
The AI maintainer team includes models from Anthropic (Claude Opus 5, Claude Fable 5), Google (Gemini 3.1 Pro), OpenAI (GPT-5.6 Sol / Codex), and Moonshot (Kimi K3).
Note su licenza e rischi
MIT License.
Editorial verification 2026-08-09: repository URL, owner, description, license and repository statistics were reviewed. License metadata: MIT. README was fetched for the channel draft; re-check repository dependencies, releases and model terms before production use.
Rilascio e manutenzione
As of June 2026, the canonical repo recorded 900+ merged PRs and 1,100+ closed issues. Version 13 (v13) introduces multi-tenant cloud deployment, allowing the Agent OS to be pointed at external codebases. The codebase contains roughly 191,000 lines of engine source, 306,000 lines of agent-readable cognitive content, and 36,000 lines of guides across ~7,200 files.