@claude-flow/neural weekly downloads statistics

@claude-flow/neural

Self-Optimizing Neural Architecture (SONA) for Claude Flow V3 — adaptive learning, trajectory tracking, pattern reuse, and 7 RL algorithms in a single package.

What this is

A self-contained learning module that records agent execution trajectories, distills them into reusable patterns, retrieves matches for new tasks, and adapts via SONA + LoRA + EWC++. Designed to be the substrate that the Claude Flow CLI's intelligence layer composes onto — the package owns the algorithms, the CLI owns the orchestration.

Install

Note (2026-05-16): @claude-flow/neural@3.0.0-alpha.9+ pins @ruvector/sona to the exact known-good 0.1.5 because @ruvector/sona@0.1.6 shipped as an empty publish (README + package.json only — no index.js, no native bins). Prior alpha.8 used "latest" and broke on every fresh install. The pin will stay until @ruvector/sona@0.1.7+ ships with content.

Standalone use (without the Ruflo CLI)

Quick start (recommended)

NeuralLearningSystem is the high-level entry point — it wires SONAManager, ReasoningBank, and PatternLearner together so callers don't have to:

Lower-level API: SONA Manager

For callers that want to manage trajectories and patterns directly:

Learning modes

| Mode | Adaptation | Quality | Memory | Use case |

| real-time | <0.5ms | 70%+ | 25 MB | Production, low-latency | | balanced (default) | <18ms | 75%+ | 50 MB | General purpose | | research | <100ms | 95%+ | 100 MB | Deep exploration | | edge | <1ms | 80%+ | 5 MB | Resource-constrained | | batch | <50ms | 85%+ | 75 MB | High-throughput |

ReasoningBank + PatternLearner (separately accessible)

NeuralLearningSystem composes them; you can also use them standalone:

RL algorithms (7 included)

Imports use the Algorithm suffix where applicable:

LoRA configuration

EWC++ (Elastic Weight Consolidation)

Prevents catastrophic forgetting when adapting to new domains:

Event system

Performance targets

| Metric | Target | Typical |

| Adaptation latency | <0.05 ms | 0.02 ms | | Pattern retrieval | <1 ms | 0.5 ms | | Learning step | <10 ms | 5 ms | | Quality improvement | +55% | +40–60% | | Memory overhead | <50 MB | 25–75 MB |

TypeScript types

Integration with @claude-flow/cli

The CLI's intelligence layer (hooks_intelligence_*, neural_* MCP tools, /intelligence dashboard) is the primary consumer. Phase 1 of the convergence (#1773) adds a thin bridge in cli/src/memory/neural-package-bridge.ts that lazy-loads NeuralLearningSystem so cli's intelligence handlers can call into the package surface alongside the existing local implementation. Future phases migrate cli's LocalSonaCoordinator and LocalReasoningBank to wrap this package's SONALearningEngine and ReasoningBankAdapter.

If you're building a Ruflo plugin that wants neural learning, depend on @claude-flow/neural directly rather than reaching into cli internals.

Dependencies

• @claude-flow/memory — vector memory for patterns • @ruvector/sona — SONA learning engine

Related packages

• @claude-flow/memory — memory backend • @claude-flow/cli — primary consumer + MCP tool surface • @claude-flow/cli-core — lite path (no neural; for plugin scripts)

License

MIT

Description from the npm listing.