convez.ai
MCP ServerAutomaçãoAvançado
multinex-ai

Multinex — Camada Adaptativa para Claude

multinex-ai/multinex-skills

Memória persistente, proteção de dados e workflows de agentes para Claude

Ver no GitHub
InicialPluginScore · de 0 a 100

Plataforma de infraestrutura de IA hospedada que conecta Claude a três produtos principais via MCP: MemQ (memória persistente e soberana entre sessões), Shield (proteção de dados e conformidade com detecção de injeção de prompt) e Legion (fluxos de agentes governados e auditáveis). Inclui skills prontos para handshake inicial, operações de memória, loops de aprendizado e handoff entre agentes.

Instalação

$npx convez-ai multinex-ai/multinex-skills

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
0
Forks
0
Acompanhando
0
Saúde
50%
Último commit
há 5 meses
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Manter contexto e conhecimento persistente entre sessões longas do Claude
  • Implementar pipelines multi-agente auditáveis com checkpointing de estado
  • Proteger dados sensíveis com mascaramento de PII e detecção de injeção de prompt
  • Capturar e reutilizar conhecimento compartilhado entre múltiplos assistentes

Veredito

Solução enterprise robusta para Claude com foco em memória persistente, segurança e orquestração de agentes. Requer OAuth setup e API key, adequada para equipes que precisam de governança, auditoria e contexto durável em workflows complexos.

README

A tradução para o português deste README ainda está na fila.

Multinex — Claude Adaptive Layer

Multinex is a hosted AI infrastructure platform. This package is the official Claude Code plugin and npm distribution for the Multinex adaptive layer — a single plugin that connects Claude to all Multinex products through one hosted MCP endpoint:

  • MemQ — persistent sovereign memory (the flagship). Episodic, semantic, procedural, and checkpoint storage across all sessions. Soul Journal, Brain Pathways, Mnemosyne context compression, and governed _commons shared knowledge.
  • Multinex Shield — AI data protection and compliance. PII masking, prompt injection detection, data classification, and audit trails for AI workloads.
  • Legion — governed agent workflow modules. Modular, auditable multi-agent pipelines with checkpointed state machines and enforced handoff boundaries.

Install the Claude Code plugin

/plugin marketplace add multinex-ai/multinex-skills
/plugin install multinex@multinex

Then sign up for a free account and complete the OAuth flow:

https://billing.multinex.ai/signup

Start every working session with reunion to handshake with the MemQ runtime and restore your persistent context.


Install via npm

The npm package ships the plugin manifests and skills as installable assets. There is no runtime code — MemQ runs as a hosted service.

npm install @multinex/multinex
# or
npx @multinex/multinex

After installing, wire the plugin by pointing your Claude Code config at the installed package directory, or use the Claude Code plugin commands above.

To publish to npm (for package maintainers only — requires npm auth):

npm publish --access public

MCP endpoint (direct / raw)

Point any MCP client at the hosted endpoint with mcp-remote:

{
  "memq": {
    "command": "npx",
    "args": ["-y", "mcp-remote@latest", "https://mcp.multinex.ai/mcp/v1"],
    "env": { "MEMQ_API_KEY_AUTH_HEADER": "Bearer <your-memq-key>" }
  }
}

OAuth-capable clients can omit the key and authenticate interactively (discovery at https://billing.multinex.ai/.well-known/oauth-authorization-server).


Skills

The plugin ships six skills. Each is a SKILL.md file with YAML frontmatter that Claude Code loads automatically when the plugin is active.

SkillTrigger / Purpose
reunion-firstStart every session: REUNION handshake, namespace discovery, Mnemosyne hydration
memq-operationsFull MemQ reference: all tools, memory types, retrieval model, operating constitution
learning-loopLong-running sessions: observe, encode, replay, reinforce, consolidate
team-handoff-commonsCross-agent handoffs: retrieve shared context, write back root-cause records
multinex-overviewProduct routing: MemQ vs Shield vs Legion decision tree, MCP connection guide
commons-qualityDurable quality capture: record resolved tool/product issues in a consistent, searchable format for commons promotion

reunion-first

Invoked at the start of every Claude Code session when the Multinex plugin is active. Runs reunion → namespace_info → mnemosyne_context in order, then sets up checkpointed write-back and a session closing sequence (brain_reinforce → brain_consolidate_sleep → save_context).

memq-operations

Comprehensive operating reference for MemQ: memory type taxonomy, retrieval scoring model (vector + graph + hot bus fusion), every tool deep-referenced, quota awareness, tag taxonomy, session patterns, and the full operating constitution.

learning-loop

Models a Claude Code session as five pathways: Observe, Encode, Replay, Reinforce, Consolidate. Guides Claude through journal_record → save_context → brain_recall_episode → brain_reinforce → brain_consolidate_sleep as a structured improvement loop.

team-handoff-commons

For multi-agent and multi-session workflows. Immediate retrieval from _commons and shared namespaces on startup. Structured root-cause write-back (failure_record → decision_point → add_memory) after resolution. Team handoff write-back via plan_state_write + reflection_handoff.

multinex-overview

Routes Claude to the correct Multinex product for any task. Covers MemQ, Shield, and Legion use cases, awareness rules (e.g. never store PII in MemQ, journal Shield alerts), Legion + MemQ integration pattern (plan_state_* tools), and the full routing decision tree.

commons-quality

Enforces a consistent capture format for resolved tool/product issues so they form a searchable, governed grouping in the Soul Journal. Every record captures: tool/product used, issue, root cause, resolution, and proof. Uses journal_record(type: "failure_record") + add_memory(memory_type: "procedural") with consistent commons-quality tags. Optionally surfaces promotion candidates via commons_resonance.


MCP Registry

Published as ai.multinex/memq (see server.json). The MCP Registry manifest covers OAuth discovery, pricing, and endpoint metadata.


Signup and Docs

  • Free 15-day trial: https://billing.multinex.ai/signup
  • Dashboard + API keys: https://billing.multinex.ai/dashboard
  • Product hub: https://multinex.ai
  • MemQ: https://multinex.ai/memq
  • Shield: https://multinex.ai/shield
  • Legion: https://multinex.ai/legion
  • Status: https://status.multinex.ai

License

MIT — see LICENSE. Multinex products are hosted commercial services; the manifests and skills in this package are provided to enable connection to them.

Ver o original no GitHub

Bom para

Relacionados