Yuva AI

yuvaog.com | Turn your AI coding tool into a coordinated multi-agent system.

What It Does

Yuva AI adds structure to AI coding. Instead of your AI guessing about your codebase, it gets real context, enforced rules, and coordinated workflows.

• Neural Graph — Maps your code relationships. AI gets only the relevant context, not the whole codebase.
• Security Scanning — Catches hardcoded secrets, vulnerable deps, and config issues before they ship.
• Quality Gates — Enforces lint, tests, JSDoc, no console.log, and custom rules. Work isn't "done" until they pass.
• Prompt Enforcement — Machine-verified rules. If the AI touches protected files, its work is rejected automatically.
• Swarm Mode — Multiple AI workers (executor, tester, reviewer) coordinate through a shared task bus.
• Loop Engine — Fully autonomous: AI plans tasks, workers build, gates verify, AI reviews and replans.
• Session Persistence — Never lose context between conversations. Auto-saves after every command.

Install

npm install -g yuva-ai
cd your-project
yuva init

That's it. Open your project in your AI tool — it reads AGENTS.md and knows what to do.

Commands

# Setup
yuva init                        # Auto-detect AI tool + build neural graph
yuva doctor                      # Diagnose setup issues
yuva status                      # Project overview

# Scan
yuva scan code                   # Analyze codebase (routes, models, env vars)
yuva scan security               # Find secrets, vulnerable deps, config issues

# Graph
yuva graph build                 # Build code knowledge graph
yuva graph query "auth"          # Search for relevant code nodes
yuva graph context "fix login"   # Preview what context a task would get

# Gates
yuva gate                        # Run all quality gates (lint + test + build)
yuva gates                       # Run plugin gates (console.log, TODO, JSDoc, etc.)

# Agents
yuva agent list                  # List all 12 agents
yuva agent show <name>           # Get agent prompt
yuva agent orchestrate           # Scan project context for AI

# Swarm (multi-worker)
yuva swarm init                  # Create task bus
yuva swarm plan "build auth"     # Break goal into tasks
yuva swarm spawn                 # Open worker terminals
yuva swarm start                 # Orchestrator dashboard
yuva task add "title" --role executor
yuva worker next --role executor
yuva task done <id> --summary "..."

# Loop (fully autonomous)
yuva loop run "add auth with tests"   # Plans, builds, verifies, replans
yuva loop stop                        # Stop the loop

# Session
yuva session start "goal"        # Start tracking
yuva session resume              # Get full context
yuva session end                 # End session

# Cost
yuva cost                        # Show AI usage
yuva cost set-budget 50          # Set spending limit

Agents

| Agent | Purpose | |-------|---------| | existingcode | Analyze codebase before changes | | requirements | Gather what to build | | riskassessment | Identify risks | | planning | Design architecture | | execution | Implement code | | tester | Write and run tests | | reviewer | Code quality audit | | security | Vulnerability scan | | debugger | Fix bugs | | refactor | Improve code | | continuity | Resume from last session | | statemanager | Update session state |

Works With

Commercial: Claude Code, Cursor, Windsurf, GitHub Copilot, Gemini, Codex, Amazon Q, Cody, Antigravity Open Source: Ollama, LM Studio, Jan, Continue, Aider, OpenCode, Kilo Code

yuva llm use cursor      # Switch tool
yuva llm use ollama      # Use local model

How It Works

Your AI Tool
    │
    ▼
AGENTS.md (reads this on startup)
    │
    ├── yuva agent orchestrate → project context (JSON)
    ├── yuva graph query → relevant code nodes
    ├── yuva gate → quality enforcement
    └── yuva task done → enforcement + gates + graph learning

The AI gets real project context, follows enforced rules, and only declares work done when quality gates pass.

Protected Files

These files are never modifiable by AI workers:

.yuva/  .session/  .aiautomations/  AGENTS.md  CLAUDE.md
.claude/  .cursor/  package-lock.json  yarn.lock

If the AI touches any of these, its task is automatically rejected.

Development

npm install
npm test                 # 300 tests
npm run lint

License

MIT