convez.ai
MCP ServerAutomaçãoAvançado
Who-Visions

NouGenShards — Memória Local para Agentes de IA

Who-Visions/NouGenShards

Memória persistente local com busca semântica para agentes de IA

Ver no GitHub
BomPluginScore · de 0 a 100

MCP Server que captura e indexa fragmentos de contexto ("shards") de múltiplas ferramentas de IA (Claude, Gemini, Cursor) em um banco SQLite local com FTS5. Permite que agentes recuperem soluções anteriores por relevância ponderada, eliminando re-prompts desnecessários e mantendo memória persistente sem dependência de cloud.

Instalação

$npx convez-ai Who-Visions/NouGenShards

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
5
Forks
2
Acompanhando
1
Saúde
75%
Último commit
há 2 dias
Versão
supervisor-v1.3.2
Issues abertas
18
Contribuidores
2

Casos de uso

  • Agentes de coding que precisam reutilizar fixes e decisões anteriores
  • Equipes que consolidam conhecimento espalhado em múltiplas plataformas de IA
  • Desenvolvimento offline com índices locais de boas práticas comprovadas

Veredito

Ferramenta sólida para contornar amnésia de sessão em agentes, com arquitetura bem pensada (sharding, BM25 + semântico, outcome-weighting). Licença source-available (não aberta) é clara; implementação em Python/SQLite garante portabilidade. Ideal para quem já trabalha com múltiplos LLMs e quer federalizar aprendizado.

README

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


title: NouGenShards Node emoji: 🧠 colorFrom: indigo colorTo: purple sdk: docker app_port: 7860 pinned: false

"Nou Gen" means "We have" in Haitian Creole. NouGenShards means: We have memory. 🇭🇹 Built by Who Visions to empower global diaspora intelligence.

Every AI tool you use forgets when the session ends. NouGenShards scans your machine for scattered AI traces — Claude, Gemini, Cursor, Codex, and more — extracts the useful context, and stores it locally in encrypted SQLite databases you control. No cloud required.

⚠️ Source-Available, Not Open Source: This project is provided so users can inspect, learn, and trust the local client. Commercial reuse, redistribution for a fee, and competing hosted services are strictly prohibited. See LICENSE.md.


📖 CLI Workflow Example

# 1. Discover and import scattered AI history from Claude, Gemini, Cursor, etc.
$ nougen brain scan
Found:
✓ Claude (3,412 items)
✓ Gemini (1,209 items)
✓ Cursor (4,891 items)

# 2. Search your memory substrate semantically
$ nougen search "React auth bug" --semantic
[Shard #142] (8 months ago via Claude)
"Fixed JWT token expiration handler..."

🏗️ Architecture

graph TD
    A[Claude / Gemini / Cursor / Logs] -->|nougen brain scan| B[AI Memory Recon]
    B -->|Extract & Normalize| C[Normalizer]
    C -->|Deterministic Hash Routing| D[(9-Db Shard Grid)]
    D -->|BM25 + Semantic Search| E[Relevance Ranker]
    E -->|Prior/Utility Updates| F[Search API]
    F -->|Telemetry / CLI| G[Cortex HUD / TUI]

🚀 Why NouGenShards?

  • AI Memory Recon: Run nougen brain scan to discover and import your fragmented AI history across 15+ known tool formats.
  • Cortex HUD: See your memory grow — a 3x3 substrate map, high-velocity timelines, and a point-and-click shard browser. Ships as a native desktop app (Tauri) and web view.
  • Privacy First: Your core memory stays on your machine in local SQLite databases. Secrets are redacted on import, and the credential vault encrypts values at rest. Cloud platforms forget, but local memory belongs to you.
  • Relevance ranking that learns: Mark a shard as helpful and it ranks higher next time. Results are scored by a weighted blend of keyword match (BM25), semantic similarity, and your usefulness votes.
  • One search across sources: Search your local memory and any cloud nodes you connect. Results are merged into one ranked list.
  • Memory consolidation (experimental): The engine ages out stale memory over time so the most useful context stays on top.
  • Bring your own provider: Route requests through OpenRouter with automatic fallback if a model is unavailable.
  • How it works: How the memory engine maps onto the code.

📦 Quick Start

1. Install

Windows (One-Click) 🪟

# Just run the launcher
nougen.bat

Other Platforms 🐍🟢

# Using Python
pip install .

1b. Desktop HUD (Tauri)

The Cortex HUD also ships as a native desktop app (Rust + Tauri v2, React frontend):

npm install          # frontend deps
npm run tauri dev    # live-reload development window
npm run tauri build  # production app at src-tauri/target/release/

Prerequisites: Node 20+, Rust toolchain (winget install Rustlang.Rustup), and npm i -g @tauri-apps/cli. On first checkout run tauri icon src-tauri/icons/icon.png to regenerate the platform icon binaries (they are not committed). The HUD talks to the Python engine through Tauri commands (search_shards, engine_status, memory_stats) that proxy the nougen … --json CLI contract.

2. Find Your AI Brain

# Discover local AI tool history
nougen brain scan

# Import history into your local memory (dry-run by default)
nougen brain import

# Write to the database
nougen brain import --confirm

3. Check Health

nougen doctor

💾 Core Workflow

Capture Experience

nougen add "Fixed the N+1 query bug in the user controller" --tags rails,fix,performance

Search Memory

nougen search "N+1 query" --semantic

Close the Loop

# Tell the tool Shard #5 was helpful so it ranks higher next time
nougen mark 5 --worked

Agent Handoffs

Leave a structured note for the next coding agent when you transfer work between sessions (Gemini, Claude, Codex, local models). Captures the goal, git state, open tasks, and a free-text note. See docs/handoffs.md for the full protocol, schema, and environment overrides.

# Outgoing agent records where it left off
nougen handoff create --goal "Wire the Tauri sidecar" --message "frontend done, rust stubbed"

# Incoming agent reviews the latest open handoff...
nougen handoff read

# ...then acknowledges it (the read-back that marks it picked up)
nougen handoff ack --message "picking this up"

# See history and which handoffs are still open
nougen handoff list

☁️ Cloud & Hybrid Modes

NouGenShards supports three ways to use cloud intelligence:

  1. Local (Free): Use Ollama or LM Studio on your own machine.
  2. BYOK (Bring Your Own Key): Connect your own OpenAI, Anthropic, or OpenRouter keys.
  3. Who Visions Cloud (Pro): Access our hosted resilient brain with metered billing and managed sync.

See Cloud Modes and Licensing for details.


🧩 What's in this repo

This repository is the public client: the CLI, the local memory engine, bring-your-own-key adapters, AI Memory Recon, and the plugin interfaces. Some hosted and advanced features are not part of this repository.


🥇 Standards

  • ✅ 100% pass rate on 190+ unit tests.
  • 💻 Hardened for Windows, macOS, and Linux.

📜 Notice

Copyright © 2026 Who Visions LLC. All rights reserved. 🛡️ This source code is provided for visibility and personal use only. Commercial reuse is not granted.

Ver o original no GitHub

Bom para

Relacionados