Skip to content

How It Works

Nova AI transforms natural language into production-ready code through coordinated AI agents.

The Orchestration Flow

sequenceDiagram
    participant You
    participant Orchestrator
    participant KB as Knowledge Base
    participant Agents
    participant Gates as Quality Gates

    You->>Orchestrator: /novaai "add feature"
    Orchestrator->>KB: Search patterns
    KB-->>Orchestrator: Relevant standards
    Orchestrator->>You: Clarifying questions
    You->>Orchestrator: Answers
    Orchestrator->>Agents: Dispatch tasks
    Agents->>Gates: Code for review
    Gates-->>Agents: Feedback
    Agents->>You: Ready to commit

Core Concepts

1. Natural Language Input

Describe what you want in plain English:

/novaai add user authentication with JWT and refresh tokens

Nova AI understands: - Intent: What you're trying to accomplish - Context: Your codebase patterns and conventions - Constraints: Security requirements, testing needs

Before implementation, Nova AI searches a 5.4GB knowledge base for:

  • Coding standards and patterns
  • Security best practices
  • Framework-specific conventions
  • Your project's existing patterns

3. Multi-Agent Coordination

Six specialized agents work together:

Agent Role Model
Orchestrator Plans and coordinates Opus 4.5
Implementer Writes code Sonnet 4.5
Code-Reviewer Security & quality Opus 4.5
Validator Runs tests Sonnet 4.5
Architect Design decisions Opus 4.5
Linter Style fixes Haiku 4.5

4. Quality Gates

Every change passes through three gates:

flowchart LR
    A[Code] --> B[Linter]
    B --> C[Code Review]
    C --> D[Tests]
    D --> E[Commit]

    B -.->|fail| F[Fix]
    C -.->|reject| F
    D -.->|fail| F
    F --> A

Execution Modes

Interactive Mode

/novaai implement feature X
  • Asks clarifying questions
  • Shows implementation plan
  • Waits for your approval
  • Full quality gates

Headless Mode

python scripts/nova_headless.py "task" --output-format json
  • No user interaction
  • Auto-approves all steps
  • Returns structured output
  • For CI/CD pipelines

What's Next?