The AI skill that reasons before it designs.
Forma is an open-source Design Intelligence Framework for AI coding assistants. It replaces basic "generate UI" prompts with an enterprise-grade, multi-agent reasoning architecture.
- The Vision
- Core Features
- System Architecture
- Quick Start
- Usage Examples
- CLI Reference
- Compatibility
- Security & Privacy
- Contributing
Every AI coding assistant today answers the same question: > "How should this look?"
Color palettes. Font pairings. Useful—but fundamentally shallow. Real, industry-grade design requires asking harder questions before a single line of code is written:
- Who is this for, and what do they actually need?
- What does this specific industry strictly forbid?
- What cognitive psychology drives this user's decisions?
- Will this actually convert?
Forma is built to answer all of them. Automatically. In sequence.
Forma operates at the absolute pinnacle of AI software engineering. It is not just a prompt file—it is a secure, multi-language execution environment powered by Python and Go.
Before any UI is generated, Forma forces your AI through a sequential gauntlet. It analyzes the UX Pre-Flight, queries Industry Intelligence, cross-references Cognitive Psychology, applies Component Best Practices, and performs an Accessibility Audit. Only then does it write code.
Instead of forcing the AI to waste thousands of tokens rewriting an entire React component just to fix a missing aria-label, the Forma Critic agent triggers the Go-based Auto-Healer (forma heal <file>). The CLI physically parses your frontend AST and surgically injects missing ARIA labels, alt texts, and SVG properties in milliseconds.
Every generated design system is backed by advanced mathematics (OKLCH Color Spaces, Fluid clamp() Typography, and Spring Physics) and exported directly into the global W3C Design Token format.
Forma will never catastrophically crash on a developer's machine. The forma doctor command instantly diagnoses your environment. If it detects a system failure, it triggers a Graceful Halt rather than hallucinating bad code.
Forma is powered by a dual-engine architecture: a high-speed Go CLI that manages system states, and a Python Intelligence Engine that processes complex reasoning matrices.
graph TD
A[User Prompt in IDE] --> B{Forma Agent}
B -->|Calls| C[Python Intelligence Engine]
C --> D[12-Layer Reasoning Matrix]
D --> E[Industry Rulesets]
D --> F[Psychology Triggers]
D --> G[Accessibility Audit]
G --> B
B -->|Generates| H[Frontend Code]
H --> I[Forma Critic]
I -->|Detects WCAG issues| J[Go CLI: forma heal]
J -->|AST Injection| H
classDef primary fill:#6366F1,stroke:#fff,stroke-width:2px,color:#fff;
classDef secondary fill:#8B5CF6,stroke:#fff,stroke-width:2px,color:#fff;
classDef engine fill:#10B981,stroke:#fff,stroke-width:2px,color:#fff;
class B primary;
class C engine;
class D,E,F,G secondary;
Forma is built on a foundation of peer-reviewed UX research, mathematical color spaces, and cognitive psychology. It is not an LLM hallucinating designs; it is an LLM querying a strict, deterministic database.
When you prompt your IDE, Forma executes a massive background sequence before writing any code:
sequenceDiagram
participant Dev as IDE Developer
participant Forma as Forma Engine
participant DB as Knowledge Matrix
participant Output as Generated UI
Dev->>Forma: "Build a high-converting pricing table"
activate Forma
Forma->>DB: Query Psychology Triggers (SaaS)
DB-->>Forma: Inject: [Decoy Effect, Anchoring]
Forma->>DB: Query Accessibility (WCAG 2.2 AAA)
DB-->>Forma: Enforce: [Contrast > 7:1, aria-labelledby]
Forma->>DB: Map Mathematical Tokens
DB-->>Forma: Export: [OKLCH Fluid Space]
Forma-->>Output: Compile Validated UI Component
deactivate Forma
Output-->>Dev: Production-Ready Code
Forma maps UI component generation directly to established psychological triggers to maximize user retention, trust, and conversion rates.
| Cognitive Trigger | UI Execution Strategy | Primary Industry |
|---|---|---|
| Hick's Law | Progressive disclosure & automated layout reduction | SaaS, E-Commerce |
| Zeigarnik Effect | Step-based visual progress rings & incomplete states | EdTech, Gamification |
| Von Restorff Effect | OKLCH chroma-isolation for primary Call-to-Actions | Fintech, Marketing |
| Cognitive Load Theory | Fluid whitespace scaling via Golden Ratio algorithms | Healthcare, Dashboards |
Requirements: Node.js 18+ · Python 3.9+ · A supported AI assistant
npm install -g @foysalzihak/forma-cliforma init --ai claude # Claude Code
forma init --ai cursor # Cursor
forma init --ai windsurf # Windsurf
forma init --ai all # Install everywhereOnce installed, Forma operates silently in the background. Just talk to your AI naturally. Forma intercepts the visual requests and applies its reasoning layers.
Example 1: High-Converting SaaS
"Build a pricing page for our B2B SaaS. Ensure it uses the 'Scarcity' psychology trigger and passes WCAG AAA contrast."
Example 2: Industry-Specific Design
"Create a checkout flow for a Fintech app. Forma, check your industry rules for traditional banking before generating."
Example 3: Mathematical Redesign
"Refactor this dashboard. Generate a fluid typography scale using
clamp()and an OKLCH color system."
The forma-cli binary is lightning fast and handles all environment integrations.
| Command | Description |
|---|---|
forma doctor |
Runs a system diagnostic on your Python and Go environments. |
forma heal <file> |
Triggers the AST Auto-Healer to physically inject missing WCAG code. |
forma export --format tailwind |
Exports the W3C tokens into your Tailwind configuration. |
forma update |
Instantly upgrades the CLI to the latest Vanguard release. |
forma remove |
Safely purges Forma agents from your codebase. |
| Agent Platform | Support | Installation Command |
|---|---|---|
 |
✅ | forma init --ai agent |
 |
✅ | forma init --ai augment |
 |
✅ | forma init --ai claude |
 |
✅ | forma init --ai codebuddy |
 |
✅ | forma init --ai codex |
 |
✅ | forma init --ai continue |
 |
✅ | forma init --ai copilot |
 |
✅ | forma init --ai cursor |
 |
✅ | forma init --ai droid |
 |
✅ | forma init --ai gemini |
 |
✅ | forma init --ai kilocode |
 |
✅ | forma init --ai kiro |
 |
✅ | forma init --ai opencode |
 |
✅ | forma init --ai qoder |
 |
✅ | forma init --ai roocode |
 |
✅ | forma init --ai trae |
 |
✅ | forma init --ai warp |
 |
✅ | forma init --ai windsurf |
Forma is universally adaptable. Just mention your stack in the prompt, or let it intelligently default to HTML + Tailwind CSS.
| Icon | Framework | Domain |
|---|---|---|
| React | Web UI Library | |
| Next.js | Web Fullstack Framework | |
| Vue.js | Web UI Library | |
| Nuxt.js | Web Fullstack Framework | |
| Svelte | Web Compiler | |
| Astro | Static Site Generator | |
| Angular | Web Framework | |
| SolidJS | Web UI Library | |
| Remix | Web Fullstack Framework | |
| Tailwind CSS | Global Styling System | |
| Bootstrap | CSS Framework | |
| Material UI | Component Library | |
| Flutter | Cross-Platform Mobile | |
| React Native | Cross-Platform Mobile | |
| SwiftUI | iOS Native | |
| Jetpack Compose | Android Native | |
| Laravel | PHP Fullstack Framework | |
| Django | Python Fullstack Framework | |
| Ruby on Rails | Ruby Fullstack Framework |
Forma is built for Enterprise, which means privacy is non-negotiable.
- Zero Telemetry: Forma does not track your usage, collect your prompts, or phone home.
- Local Execution: The Go CLI and Python Engine run entirely on your local machine.
- Sandboxed Agent Logic: All AI requests are handled exclusively by your IDE's existing LLM connections. Forma acts merely as local context and tooling.
Forma deeply respects your codebase. If you want to remove the AI logic from your project, you don't have to hunt down hidden files.
# 1. Safely purge all Forma agents from your repository
forma remove
# 2. Uninstall the global binary
npm uninstall -g @foysalzihak/forma-cliWe welcome contributions from the community! Forma is a dual-engine framework built with Go (CLI) and Python (Intelligence).
# Clone the repository
git clone https://github.com/fzihak/forma.git
cd forma
# Run the Automated Test Suite
cd forma-cli
go test ./internal/healer/...
python -m unittest ../src/tests/test_engine.pyPlease see CONTRIBUTING.md for full architectural guidelines.
Built with Go, Python, and Mathematical Neuro-Design.
MIT License © Foysal Zihak