sairaman436/vybe-intelligence-vault: vybe-intelligence-vault

📌 Overview

Most AI knowledge bases go stale the moment you stop updating them. Vybe Intelligence Vault doesn't — it runs itself.

A GitHub Actions pipeline wakes up every 3 hours, discovers emerging AI/ML resources, evaluates them with an LLM scoring engine, and commits the ranked results back into the repo. No human in the loop. No manual curation.

The result: a self-reinforcing knowledge graph of 5,928 indexed resources spanning AI agents, RAG architectures, MCP servers, and modern web tooling — always current, always queryable by local agents via an HTTP gateway.

Built for: AI engineers who want a living knowledge base they can plug into agentic workflows, not a static awesome-list that someone forked two years ago.

⚙️ How It Works

Every 3 hours:

  GitHub Actions Cron
       │
       ▼
  evaluate_repo.py          ← discovers candidate resources from configured topics
       │
       ▼
  LLM Scoring Engine        ← qwen2.5:14b (local) or cloud fallback
       │  scores: quality, rag_relevance, tech_stack match
       ▼
  vault-core/               ← ranked .md files committed back to repo
       │
       ├─▶ rebuild-index.yml  ← triggers on push
       │         │
       │         ▼
       │   nomic-embed-text   ← local Ollama embeddings
       │         │
       │         ▼
       │   vault-index.json   ← semantic node/edge graph (cosine sim > 0.75)
       │         │
       │         ▼
       │   React 3D Map       ← WebGL intelligence visualization (SWR polling)
       │
       └─▶ Orchestrator :3456 ← MCP gateway for agent context injection

Scoring Pipeline

Each resource is evaluated across 4 dimensions:

Semantic Graph

Embeddings via nomic-embed-text (Ollama). Two nodes are linked if:

• Cosine similarity > 0.75 → similar_to
• Shared tech stack → depends_on
• Same category + shared tags → references

Edge weight = cosine_sim + (shared_tags × 0.08)

MCP Gateway

HTTP bridge on :3456. Send a vault file path → receive a clean, LLM-formatted context block. Agents can pull any resource into their context window without reading the filesystem directly.

# Example agent request
curl -X POST http://localhost:3456/inject \
  -d '{"path": "ai/agents/tool-use-patterns.md"}'

🏗 Architecture

graph TD
    A[⏰ Cron / Dispatch Trigger] -->|every 3h| B(evaluate_repo.py)
    B -->|LLM score| C{Decision Engine}
    C -->|pass threshold| D[vault-core/]
    C -->|reject| X[❌ Discarded]
    D -->|git push| E(rebuild-index.yml)
    E -->|nomic-embed-text| F[vault-index.json]
    F -->|SWR poll| G[���� React 3D Map]
    H[🤖 AI Agent] -->|MCP request| I[Orchestrator :3456]
    I -->|read + format| D

    style A fill:#1f2937,color:#e5e7eb
    style D fill:#111827,color:#e5e7eb
    style G fill:#1f2937,color:#e5e7eb
    style H fill:#374151,color:#e5e7eb

Key Design Decisions

Write-lock state management — state.lock prevents collision writes when multiple pipeline jobs run concurrently. All mutations are append-only to vault-events.log (JSONL). In-memory reads use a 30s TTL. → scripts/state-manager.js

Hybrid inference — Pipeline tries local Ollama first (zero cost, no rate limits). Falls back to cloud LLM if Ollama is unavailable. Scoring is deterministic via fixed seed. → scripts/evaluate_repo.py

Bot commits on heatmap — Git identity configured so automated commits register on the contribution graph. Pipeline runs as vybe-bot with a PAT scoped to repo only. → .github/workflows/harvester.yml

🚀 Quick Start

Prerequisites

• Ollama installed and running
• Node.js 18+
• Python 3.10+

Setup

# Clone
git clone https://github.com/sairaman436/vybe-intelligence-vault.git
cd vybe-intelligence-vault

# Pull required models
ollama pull nomic-embed-text   # embeddings
ollama pull qwen2.5:14b        # scoring

# Install dependencies
npm install
pip install -r requirements.txt

# Start everything (MCP server + orchestrator + web UI)
bash scripts/vault-init.sh

Verify

# Check service health, ports, and event log
bash scripts/vault-status.sh

Open http://localhost:3000 for the 3D intelligence map.

Configure Topics

Edit vault-core/config.yaml to control what gets harvested:

topics:
  - ai-agents
  - rag-architectures
  - mcp-servers
  - llm-inference
  - next-gen-web

token_budget: 4096
score_threshold: 0.65

📈 Trending Signals

Top rising resources based on momentum and community velocity.

• 🔼 Medium Members Can Listen To Any Medium Story With The Speechify Play Button | Speechify • Rank: +1
• 🔼 Home - The GitHub Blog • Rank: +1
• 🔼 LinkedIn • Rank: +1
• 🔼 GitHub Changelog • Rank: +1
• 🔼 The Agent Skills Directory • Rank: +1

🌟 New Discoveries

Fresh intelligence recently indexed into the vault.

• 🆕 EngineeringBringing more control over your connectorsJune 24, 2026By Mistral AI • Score: 0
• 🆕 langchain-ai/langgraph • Score: 0
• 🆕 ProductIntroducing Search ToolkitProduction search pipelines, anywhere.May 28, 2026By Mistral • Score: 0
• 🆕 CompanyAI Now Summit 2026Innovations for global enterprises solving the world’s hardest problems. May 28, 2026By Mistral • Score: 0

💤 Recently Inactive

Resources showing declined activity or relevance.

• None.

The stats shown here are generated from the current vault content. They refresh automatically when the bot finds changes.

📁 Repository Layout

vybe-intelligence-vault/
├── .github/
│   └── workflows/
│       ├── harvester.yml        # Main 1h cron pipeline
│       └── rebuild-index.yml    # Triggered on vault-core/ push
│
├── vault-core/
│   ├── config.yaml              # Topics, token budgets, score thresholds
│   ├── vault-index.json         # Compiled semantic node/edge graph
│   └── vault-events.log         # Append-only JSONL event ledger
│
├── intelligence-map/            # React 19 + WebGL 3D dashboard
│
├── mcp-server/                  # FastMCP integration server
│
├── scripts/
│   ├── evaluate_repo.py         # Resource discovery + LLM scoring
│   ├── orchestrator/
│   │   └── context-injector.js  # MCP context formatter
│   ├── state-manager.js         # Lock-safe state writes
│   ├── build-index.js           # Embedding + edge compiler
│   ├── vault-init.sh            # Startup daemon (concurrent)
│   └── vault-status.sh          # Port + health diagnostics
���
├── ai/                          # Indexed resources by category
│   ├── agents/
│   ├── rag/
│   ├── models/
│   └── mcp/
│
└── search-index.md              # Flat searchable index

🗺 Roadmap

• Vector search API over vault-index.json (FastAPI endpoint)
• Discord/Slack bot that answers "what's new in RAG this week?"
• Contributor scoring — track who surfaces the highest-value resources
• Export to Obsidian vault format
• GitHub App so others can run their own vault instance

🤝 Contributing

PRs welcome. Read CONTRIBUTING.md first.

# Run the scoring pipeline locally against a single URL
python scripts/evaluate_repo.py --url https://github.com/your/repo --dry-run

Bug reports → open an issue MIT License — see LICENSE