Kaide

AI-native architecture kit for modern React. High-discipline protocols for autonomous agents.

Kaide (Etymology: Turkish): A fundamental rule, principle, or base that provides structural integrity. The name cleverly embeds AI (Kaide), reflecting its core mission.

Getting Started

Installation & Setup

1. Run the CLI to automatically install the configuration files:

   npx kaide init

2. (Recommended) For maximum adherence, point your AI tool to your identity in Settings/Project Rules:

Adhere to AGENTS.md for persona/planning and docs/ for architectural constitutions.

Usage Workflow

This kit follows a Plan-First Protocol:

• Initialize: Ask the AI to read the rules: Read all .mdc files and AGENTS.md.
• Plan: Propose a feature: I want to build a [Feature Name]. Create an architectural plan.
• Execute: After approving the plan, the AI will generate code following the strict constraints of the relevant .mdc files.

The Problem We Solve

Kaide provides the missing governance layer to turn AI from a simple coder into a high-level Staff Engineer.

Without a structured architectural constitution, AI agents often suffer from:

• Hallucinations & Drift: Generating outdated patterns or mixing incompatible framework logics.
• Context Bloat: Wasting tokens and losing focus by re-learning project rules in every session.
• Architectural Erosion: Degrading code quality over time due to a lack of enforced modularity.
• Inefficient Execution: Writing boilerplate-heavy code instead of precise, feature-driven logic.

Core Architecture

The system operates on a Three-Layer Governance model:

• Global Constitution (core-principles.mdc): Universal rules for code integrity, naming, and error management.
• Framework Adapters (nextjs.mdc, tanstack-start.mdc): Optimized protocols for Next.js 16 and TanStack Start, ensuring the AI utilizes the best patterns for each.
• Technical Standards (api.mdc, forms.mdc, testing.mdc, etc.): Domain-specific constraints for state, validation, and performance.

Key Features

• Elite AI Persona: Integrated AGENTS.md to transform your LLM into a specialized Staff Architect.
• Feature-Driven Design: Everything is organized by domain in src/features/.
• Zod-First Type Safety: End-to-end type safety derived from schemas.
• Optimized for Modern Tech: Seamless support for Next.js and TanStack Start for world-class web applications.
• Performance-First: Built-in rules for LCP optimization, CLS prevention, and maximum efficiency.

LLM Agnostic Usage

While optimized for Cursor (.mdc), Kaide is built on pure Markdown and works with any "Agentic" AI (Claude Code, Windsurf, GitHub Copilot CLI, etc.):

• All .mdc files are standard Markdown. No conversion is needed.
• Tip: Simply tell your agent: "Read all rules in .cursor/rules/ and follow the AGENTS.md persona" at the start of your session.

Technical Standards

Layer	Tech Stack	Tooling
Server State	TanStack Query	tanstack-query.mdc
Client State	Zustand	state-management.mdc
Validation	Zod	typescript.mdc
Testing	Vitest + RTL	testing.mdc
Styling	Tailwind CSS	ui-components.mdc

Security & Performance

• Zero Inline Script: Strictly enforced CSP compliance.
• Locked Boundaries: Server-only logic is isolated from client bundles.
• Optimized Assets: Automatic enforcement of modern formats and performance patterns for peak scores.