Skip to content

README.md

Latest commit

 

History

History
82 lines (59 loc) · 3.31 KB

README.md

File metadata and controls

82 lines (59 loc) · 3.31 KB

Development Environment

Basic Usage

  1. Install uv and dagger
  2. Make sure you have ANTHROPIC_API_KEY and GEMINI_API_KEY env vars available.
  3. Run the commands

Commands

All the commands are run using uv from agent directory.

uv run test  # run all tests
uv run test_e2e  # only run e2e test
uv run lint   # lint and autofix the code
uv run update_cache  # update the LLM cache, required for new prompts or generation logic changes
uv run generate "my app description" # generate an app from scratch using full pipeline, similar to e2e test
uv run interactive  # a naive debug client working with local server

App templates

We support three app templates:

  • trpc: Full-stack web app with Bun, React, Vite, Fastify, tRPC and Drizzle
  • python: Data-oriented app with Python, NiceGUI, and SQLModel
  • laravel: Full-stack web app with Laravel, React, TypeScript, Tailwind CSS, and Inertia.js (early version)

You can specify the template when generating an app, e.g.:

uv run generate "make a dashboard showing my current stock portfolio value using up to date prices from yfinance" --template_id "nicegui_agent"

or

uv run generate "make a presentation-like website with multiple pages with content and next buttons on each. last page mush show a counter that increments each time this presentation has been shown" --template_id "laravel_agent"

Architecture

This agent doesn't generate entire applications at once. Instead, it breaks down app creation into small, well-scoped tasks that run in isolated sandboxes:

tRPC Applications

  1. Database schema generation - Creates typed database models
  2. API handler logic - Builds validated Fastify routes
  3. Frontend components - Generates React UI with proper typing

Laravel Applications

  1. Database migrations & models - Creates Laravel migrations with proper syntax and Eloquent models with PHPDoc annotations
  2. Controllers & routes - Builds RESTful controllers with Form Request validation
  3. Inertia.js pages - Generates React components with TypeScript interfaces
  4. Validation & testing - Runs PHPStan, architecture tests, and feature tests

Each task is validated independently using language-specific tools (ESLint/TypeScript for JS, PHPStan for PHP), test execution, and runtime logs before being accepted.

More details on the architecture can be found in the blog on our design decisions.

Custom LLM Configuration

Override default models using backend:model format:

# Local (Ollama and LMStudio supported)
LLM_BEST_CODING_MODEL=ollama:devstral
LLM_UNIVERSAL_MODEL=lmstudio:[host] # just lmstudio: works too

# Cloud providers
OPENROUTER_API_KEY=your-key
LLM_BEST_CODING_MODEL=openrouter:deepseek/deepseek-coder

Among cloud providers, we support Gemini, Anthropic, OpenAI, and OpenRouter.

Defaults:

LLM_BEST_CODING_MODEL=anthropic:claude-sonnet-4-20250514   # code generation
LLM_UNIVERSAL_MODEL=gemini:gemini-2.5-flash-preview-05-20  # universal model, chat with user
LLM_ULTRA_FAST_MODEL=gemini:gemini-2.5-flash-lite-preview-06-17  # commit generation etc.
LLM_VISION_MODEL=gemini:gemini-2.5-flash-lite-preview-06-17  # vision model for UI validation