AI Driven App Test Platform MVP

A deployable MVP for AI-assisted mobile app testing. It uses Maestro as the first automation executor while keeping an internal Test Case DSL, so Appium, Playwright, or device-cloud executors can be added later.

Display

Capabilities

• Upload Figma MCP context and Figma design images
• Retrieve related context with the built-in lightweight RAG layer
• Generate structured test cases and persist them in SQLite
• Convert internal Test Case DSL into Maestro YAML flows
• Select regression cases from changed features, screens, and change summaries
• Run in dry-run mode or call the local Maestro CLI
• Generate HTML test reports
• Provide a built-in Web UI and REST API

Quick Start

Recommended npm entrypoint:

npm run setup
npm run dev

Open:

http://127.0.0.1:8080

You can also run the Python server directly:

python3 -m ai_test_platform.server

Default database file:

./data/app.db

Design

See DESIGN.md for the current workflow, architecture, data model, RAG approach, Maestro execution path, and project memory design.

Docker

docker build -t ai-app-test-platform .
docker run --rm -p 8080:8080 -v "$PWD/data:/app/data" ai-app-test-platform

Maestro Execution

The default execution mode is dry-run, so the platform can run even without a device or Maestro CLI.

If Maestro is installed locally and you want real execution:

npm run dev:maestro

The runner writes flows to:

./data/maestro_flows

Then it attempts to run:

maestro test <flow-file>

Maestro CLI is an optional dependency. Official installation options include:

curl -fsSL "https://get.maestro.mobile.dev" | bash

On macOS, Homebrew is also available:

brew tap mobile-dev-inc/tap
brew install mobile-dev-inc/tap/maestro

Real Maestro execution also requires Java 17+ and a running Android Emulator, iOS Simulator, or connected device.

Dependencies

This MVP has no third-party Python dependency. The HTTP server, SQLite storage, lightweight RAG layer, and AI provider calls are implemented with the Python standard library, so npm install does not download any runtime packages.

Figma-Only MVP Mode

The current MVP intentionally focuses on Figma-driven testing. PRD ingestion is hidden from the UI and the generator no longer creates PRD/Figma alignment cases.

Recommended flow:

1. Use Cursor or another MCP-capable tool to connect to Figma MCP.
2. Export one or more Figma design screens as PNG/JPG/WebP.
3. Upload all design images in the Web UI.
4. If an AI provider is configured, uploaded Figma images are parsed by the configured vision-capable model into source models.
5. Generate test cases from the Figma design context.
6. Generate Maestro flows and run them with npm run dev:maestro.

Without an AI provider, uploaded images are stored as AI-ready artifacts and the system falls back to rule-based generation.

LlamaIndex, pgvector, and LangGraph are production-upgrade recommendations, not required by the current MVP. For a production version, a likely stack is:

FastAPI + LlamaIndex + PostgreSQL/pgvector + LangGraph/Temporal

AI Configuration

The platform uses real AI only when a provider is configured. Without it, test case generation falls back to the deterministic rule-based generator.

You can configure the provider either through environment variables or from the Web UI under AI Platform Settings. UI-entered API keys are kept only in the running server process and are not returned by the API.

Supported provider modes:

• openai: OpenAI Responses API with strict JSON Schema output
• compatible: OpenAI-compatible Chat Completions API
• ollama: local Ollama-style chat API
• disabled: force rule-based fallback

export AI_PROVIDER=openai
export OPENAI_API_KEY=...
export AI_MODEL=gpt-4.1-mini
npm run dev

OpenAI-compatible provider example:

export AI_PROVIDER=compatible
export AI_BASE_URL=http://127.0.0.1:8000/v1
export AI_MODEL=your-model
export AI_API_KEY=optional-key
export AI_RESPONSE_FORMAT=json_object # use none if your gateway rejects response_format
npm run dev

Ollama-style local provider example:

export AI_PROVIDER=ollama
export AI_BASE_URL=http://127.0.0.1:11434
export AI_MODEL=llama3.2
npm run dev

Current AI usage:

• Provider facade over OpenAI, OpenAI-compatible, and Ollama-style APIs
• Structured JSON output with provider-specific enforcement where available
• RAG context + Figma/design context -> structured Test Case DSL

Status endpoint:

GET /api/ai/status

API Summary

• GET /api/health
• GET /api/ai/status
• GET /api/ai/config
• POST /api/ai/config
• POST /api/documents
• GET /api/documents
• POST /api/source-files
• GET /api/source-files
• POST /api/figma/mcp-context
• GET /api/figma/artifacts
• POST /api/generate-cases
• GET /api/test-cases
• GET /api/memory
• GET /api/memory/context
• POST /api/source-models
• GET /api/source-models
• POST /api/change-sets
• POST /api/case-suggestions
• POST /api/test-cases/{id}/approve
• POST /api/test-cases/{id}/maestro
• POST /api/regression/select
• POST /api/runs
• GET /api/runs/{id}
• GET /api/reports/{run_id}.html

Thanks

Maestro

Codex