LLM-powered structured prompts and transformations and other random wackiness for video generation models

• Multi-Provider Architecture: Use any LLM (OpenRouter 100+ models, OpenAI, Gemini, local servers) for any task
• Intermediate-First Generation: Create model-agnostic semantic prompts, instantly export to Sora 2, Veo 3, or Generic formats
• Task-Level Granularity: Assign different providers/models per task (generate, mix, transform, etc.)
• Streaming Support: Live token streaming with tok/s metrics, cancel generation mid-stream
• Multi-Turn Conversations: Refine outputs iteratively with conversation tracking
• Token Tracking: Session-based usage dashboard with provider/task breakdown, export CSV/JSON
• Generate: Natural language → structured output (YAML, JSON, XML, Markdown, Natural Language)
• Transform: Convert between formats with mix options (reverse, compress, expand, technical, custom)
• Mix: Blend multiple prompts into hybrids
• Library: Local prompt management with search, favorites, versions
• Media: Image/video conditioning via vision API
• Transparency: Preview and edit all prompts before sending, customize system prompts, view intermediate representations
• Settings: 7 tabs (Providers, Tasks, API Key, Model, System Prompts, Tokens, Data)
• Privacy: 100% client-side, encrypted API keys per provider, network monitoring, audit logs

Three-panel layout with prompt library, input controls, and structured output.

View and edit system + user prompts before sending to the API.

Upload images/videos, mix multiple prompts, and use AI vision analysis.

Choose from all available Gemini models with real-time availability.

Configure temperature, Top P, and max output tokens.

Customize all 4 core system prompts with live preview.

Manage storage and clear cached API responses.

Real-time network monitoring, audit logs, and privacy verification.

npm install
npm run dev

Open http://localhost:1847 (or custom port via npm run dev -- --port 12345)

API Key Setup: Add providers in Settings → Providers tab (keys stored encrypted with configurable TTL)

Get API keys:

• OpenRouter: OpenRouter (100+ models, single key)
• OpenAI: OpenAI Platform
• Gemini: Google AI Studio
• Local: Run heylookitsanllm or any OpenAI-compatible server

Documentation: User Guide - includes installation, features, and advanced usage

• Prompts/intermediates/versions/media/providers/tasks: IndexedDB (browser-local, DB v8)
• API keys: Encrypted per-provider storage (AES-GCM, configurable TTL, default 7 days)
• Cache & custom system prompts: localStorage (models list, API responses, edited prompts)

Multi-Provider Architecture: Task-level granularity - assign any provider/model to any task (generate, mix, transform, etc.)

Intermediate Architecture: Prompts stored as model-agnostic Markdown, transformed on-demand to any format (zero extra API calls)

Data only sent to configured provider APIs when you explicitly trigger generation/transformation.

• React 19 + TypeScript + Vite + Tailwind v4
• IndexedDB (idb library) - DB v8
• Multi-provider LLM support: OpenRouter, OpenAI, Gemini, local servers

• Node.js 18+
• Modern browser
• API key for at least one provider (OpenRouter, OpenAI, Gemini, or local server)

Apache License 2.0