convez.ai
MCP ServerBusca & PesquisaAvançado
mahendra-shah

OmniBrain: Oracle RAG Self-Hosted para Slack & Notion

mahendra-shah/org-brain

Oracle RAG self-hosted para Slack com roteamento multi-LLM e cache eficiente

Ver no GitHub
InicialPluginScore · de 0 a 100

MCP server enterprise que funciona como cérebro central da organização, integrando Slack e Notion via RAG (Retrieval-Augmented Generation) com roteamento multi-LLM inteligente. Oferece busca e síntese de informações em tempo real com failover automático entre provedores e cache otimizado para reduzir custos de tokens em até 80%.

Instalação

$npx convez-ai mahendra-shah/org-brain

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
1
Forks
1
Acompanhando
0
Saúde
28%
Último commit
há 5 meses
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Busca e síntese de informações em workspaces Slack conectados a Notion
  • Automação de respostas a perguntas organizacionais com contexto dinâmico
  • Dashboard de custo de tokens com análise comparativa entre provedores LLM

Veredito

Solução sólida para empresas que precisam de RAG on-premise com alta resiliência e otimização de custos. O roteamento multi-LLM com failover automático e cache local são diferenciadores reais, mas exige expertise em DevOps e configuração de MCP servers para produção.

README

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

OmniBrain: Self-Hosted Workspace RAG Oracle for Slack & Web

TypeScript Node.js Notion MCP Slack Bolt License: MIT

OmniBrain is an enterprise-grade, highly resilient, and cost-efficient self-hosted RAG (Retrieval-Augmented Generation) system. It connects Slack and a sleek glassmorphic Web QA UI to your Notion workspace using the official Notion Model Context Protocol (MCP) Server. OmniBrain orchestrates advanced multi-LLM routing to answer complex organization questions, manage tasks, and track projects in real-time.

Engineered for production reliability, it implements automated multi-LLM failover (DeepSeek ➡️ Gemini ➡️ Claude), subprocess supervisor watchdogs, local token-saving caching, input content safety filters, and Slack event de-duplication.


✨ Premium Capabilities

  • Router-Generator Split (Single-Turn RAG): Slashes input token costs by 80%+ and reduces RAG latency by converting multi-turn tool calling into a predictable, single-round parallel flow.
  • Active Multi-LLM Failover: Automatically switches providers (e.g., DeepSeek to Gemini or Claude) during API outages, network timeouts, or rate limits.
  • Dynamic Identity Resolution: Resolves Slack users' names and emails against workspace profiles to answer self-referential queries ("what are my tasks?") dynamically without hardcoded IDs.
  • Subprocess MCP Supervisor: Spawns and monitors the Notion MCP server over secure stdio channels with auto-restart on unexpected crashes.
  • Local Server-Side Caching: Fast 15-minute in-memory TTL caching with thread-aware history hashing (serves repeated thread queries in 0ms).
  • Interactive Cost Dashboard: Glassmorphic UI to track token usage, region costs (ap-south-1 Mumbai), and comparative savings.
  • Schema Safety Guardrails: Programmatically checks property types and blocks invalid Notion filters that trigger 400 Bad Request errors.

🚀 Architecture Blueprint

flowchart TD
    %% Node Styling Definitions
    classDef userNode fill:#3b82f6,stroke:#1d4ed8,stroke-width:2px,color:#fff;
    classDef serviceNode fill:#8b5cf6,stroke:#6d28d9,stroke-width:2px,color:#fff;
    classDef cacheNode fill:#10b981,stroke:#047857,stroke-width:2px,color:#fff;
    classDef llmNode fill:#f59e0b,stroke:#d97706,stroke-width:2px,color:#fff;
    classDef mcpNode fill:#ef4444,stroke:#b91c1c,stroke-width:2px,color:#fff;

    %% Elements
    User["User: Slack / Web UI"]
    App["OmniBrain Service"]
    Cache{"Memory Cache"}
    Router["LLM Router"]
    Notion["Notion MCP Server"]
    Gen["LLM Generator"]

    class User userNode;
    class App serviceNode;
    class Cache cacheNode;
    class Router llmNode;
    class Notion mcpNode;
    class Gen llmNode;

    %% Flow Path
    User -->|1. Submit Query| App
    App -->|2. Check Cache| Cache
    
    %% Cache Hit Pathway
    Cache -->|Cache Hit| User
    
    %% Cache Miss Pathway (Full RAG)
    Cache -->|3. Cache Miss| Router
    Router -->|4. Search Plan| App
    App -->|5. Run Parallel Queries| Notion
    Notion -->|6. Return Context| App
    
    App -->|7. Synthesize Answer| Gen
    Gen -->|8. Return Answer| App
    App -->|9. Update Cache| Cache
    App -->|10. Deliver Response| User

⚙️ Configuration Setup

Configure your environment settings using the provided template in .env.example. Copy the template to .env in the root directory:

cp .env.example .env

Critical Environment Settings

Only a few variables are required to get started. For the full list of options, see .env.example.

# Selected LLM Provider ("deepseek", "gemini", or "claude")
LLM_PROVIDER=deepseek

# API Keys (Configure keys for your active providers and fallbacks)
DEEPSEEK_API_KEY=your_deepseek_api_key
GEMINI_API_KEY=your_gemini_api_key
ANTHROPIC_API_KEY=your_anthropic_api_key

# Notion Integration
NOTION_API_TOKEN=secret_your_notion_token

📂 Project Structure

  • src/index.ts — App entry point coordinating Slack Bolt and Express.
  • src/config.ts — Validates and loads environment settings.
  • src/llm.ts — Multi-LLM client supporting DeepSeek, Gemini, and Claude with automatic failover.
  • src/mcp.ts — Monitors and supervises the Notion MCP subprocess client.
  • src/rag.ts — RAG core coordinating router-generator split, caches, and context.
  • src/server.ts — Express server hosting the API routes and Web QA Dashboard.
  • src/utils/ — Notion helpers, response formatters, and safety filters.

🛠️ Getting Started

Prerequisites

Installation & Run

  1. Install dependencies:

    npm install
    
  2. Build TypeScript files:

    npm run build
    
  3. Start the application:

    npm start
    

Once started, the console will show Notion database pre-caching, Slack Bolt connection setup, and local web dashboard availability.


🧪 Verification & QA Dashboard

Automated Tests

Run the standalone sandbox scenarios (which verify task fetching, user mapping, and search workflows offline):

npx tsx tests/run-batch-tests.ts

Local Web UI

When the server is active, visit:

  • Interactive Chat UI: http://localhost:3000 (test questions and inspect raw citations).
  • Cost Report Dashboard: http://localhost:3000/cost_report.html (compare cost savings, toggle currency exchange rates, and evaluate token models).

🛡️ Resiliency & Safeguards

  • Failure-Proof Failover: Automatic degradation to secondary LLMs if primary APIs fail, protecting critical client integrations.
  • Supervisor Watchdog: Automatically restarts the Notion MCP stdio subprocess within 3 seconds of a crash.
  • Safe Slack mrkdwn Formatting: Natively formats lists, bold text, checkboxes ([ ] ➡️ ⬜, [x] ➡️ ✅), and links into Slack-supported markdown.
Ver o original no GitHub

Bom para

Relacionados