Back to Ecosystem Pulse
ToolExperimentalMCP
neo
by neomjs
Self-evolving JavaScript runtime for stateful multi-agent swarms
JavaScript
Updated Jul 26, 2026
Share:
How It Works
Implements a self-evolving multi-agent runtime that runs cross-model swarms inside live apps using Neural Link, GraphRAG, and self-healing loops. Semantic Capability Matching Pattern Orchestrates agents, memory (long-term and scene graphs), and web-worker frontends to keep agents stateful and continuously adaptive. Notable features include Active Hybrid GraphRAG for contextual retrieval and a runtime geared toward in-app agent deployment and recovery. Dynamic Task Routing Pattern
The Value Proposition
As agents become persistent parts of user-facing apps, tracking their behavior, delegation patterns, and failure modes matters for trust and reliability. Neo.mjs gives teams a runtime where agents keep memory, recover from errors, and evolve—making it possible to gather agent trace data and long-term track records rather than one-off benchmark results. That persistent, context-aware execution is a missing piece for agent-to-agent evaluation and reputation because it produces the live interaction traces needed to assess reliability over time. Model Context Protocol (MCP)
Target Use Cases
Teams building in-browser or node-hosted multi-agent experiences that need persistent memory, contextual retrieval, and self-repair capabilities. Reflection Pattern
Use Cases
- Deploying stateful agent swarms inside web apps with long-term memory and scene graphs
- Collecting continuous agent interaction traces for post-hoc reliability and delegation analysis
- Building self-healing agent pipelines that detect failures and recover without human intervention
Topics
agent-memoryaiai-agentai-memorycontext-engineeringfrontendfrontend-runtimegraph-ragjavascriptjson+10 more
Similar Tools
autogencrewai
Keywords
multi-agentgraph-raglong-term-memorymulti-agent trustmcp