Neo.mjs is a framework for building AI-driven applications where multiple language models collaborate through shared memory and reasoning systems to inhabit and autonomously improve live applications.
The project addresses the limitation of single-agent AI systems by orchestrating a swarm of models from different providers—Claude, Gemini, GPT—that share reasoning through Active Hybrid GraphRAG and long-term memory. Rather than treating AI as a chat interface, Neo.mjs embeds agents directly into running applications through its Neural Link interface, allowing them to inspect semantic runtime state, mutate UI and data in real time, and operate as collaborators within the application itself. The swarm autonomously executes the full engineering lifecycle: ideating features, building implementations, cross-reviewing code, running DreamService cycles to adjust priorities, and closing self-healing loops where runtime failures and architectural issues feed back into memory and graph topology for continuous improvement.
Adoption suits teams building production applications that need autonomous AI engineering capabilities beyond conversational interfaces. The framework is designed to work with existing codebases regardless of the training data of the underlying models. The project explicitly positions itself against the industry pattern of deploying single agents, arguing that a coordinated swarm catches problems no individual model can identify in itself.
The project maintains an active development pace with substantial throughput in merged contributions and closed issues. The codebase is self-maintained by the swarm itself, demonstrating the framework's core capability in practice. Development spans multiple AI integration patterns including agent memory systems, knowledge graphs, semantic search, and multi-agent coordination through web workers and MCP server implementations.
