From Chaos to Canon. Locally.
Hi there! 👋 I'm building Synnia, an open-source tool to help manage the chaos of creative assets in the AI era.
Synnia is a Digital Asset Consistency Engine.
In most node-based tools (like ComfyUI), nodes are functions (do this -> do that). In Synnia, nodes are Assets (your images, your text, your characters).
You place your assets on an infinite canvas, organize them, and use "Recipe" to grow your project while keeping everything consistent.
I'm not a full-time professional developer, and I often found existing node-based tools a bit overwhelming. I started building Synnia because I wanted a workspace that felt intuitive and accessible—something with a gentle learning curve that invites you to play, rather than requiring you to study. It's designed for indie devs, writers, and world-builders who want to keep their creative universe cohesive.
A lot has happened since the initial release! Here's what's working now:
- Asset-Centric Architecture: Separated the "Data" (Asset) from the "View" (Node). Your data lives independently of the canvas.
- Graph Engine V4: A rewritten core that handles complex interactions, dragging, and connections smoothly.
- SQLite Backend: Robust local database storage with Content-Addressable Storage (CAS) for assets.
- Modular Backend: Clean Rust architecture with separated domains (assets, models, recipes).
- YAML-Based Recipes: Define workflows as declarative YAML packages with schemas, prompts, and executors.
- Recipe Package Architecture: Self-contained recipe packages with manifests, schemas, and prompt templates.
- Built-in Recipe Library: Ready-to-use recipes for creative work:
- 🎭 Storyteller - Generate complete character soul profiles
- 🎨 Art Director - Create visual blueprints and asset definitions
- 📦 Product Manager - Production manifest generation
- 🎬 Media Recipes - Image/video generation workflows
- Recipe IDE: Built-in Mini-IDE for editing and managing recipes with multi-level category navigation.
- Edge Auto-Fill: Connections automatically populate linked field values.
- Unified Model System: Single interface for multiple AI providers:
- OpenAI (GPT-4o, DALL-E 3)
- Anthropic (Claude 3.5 Sonnet)
- Google AI (Gemini 2.0, Imagen)
- FAL.ai (Flux, Kling)
- Zhipu GLM (智谱)
- OpenAI-compatible endpoints
- Multi-Turn Chat: Support for conversational recipe execution.
- Unified Executor: Standardized result handling across all providers.
- i18n Support: Full internationalization with English and Chinese.
- Deep Data Editor: Rich editor for complex nested data structures.
- MCP Bridge Integration: AI assistants can screenshot, inspect, and debug your running app via mcp-server-tauri.
- TypeScript Bindings: Auto-generated types from Rust via ts-rs.
- More Node Types: Markdown editor, JSON viewer, debug/inspection nodes.
- Recipe Sharing: Export recipes as shareable packages.
- Performance: Optimize for large canvases with many assets.
- Meta-Recipes: Compose recipes from other recipes + assets to create new workflows.
- Plugin System: Extend Synnia with community-built nodes and recipes.
- Collaboration: Real-time sync for team projects (exploring CRDTs).
- Frontend: React 19 + TypeScript + @xyflow/react + Radix UI + TailwindCSS
- Backend: Rust + Tauri 2.2 + SQLite + Actix-web
- State: Zustand + Zundo (undo/redo)
- AI: Vercel AI SDK + provider-specific SDKs
# Clone the repo
git clone https://github.com/l1veIn/Synnia.git
cd synnia
# Install dependencies
pnpm install
# Run in dev mode
pnpm tauri:devI'm still learning React, Rust, and system architecture. If you see something that makes you say "why did he do it like that?", please open an issue or a PR! I'd love to learn from you.
Built with ☕ and curiosity.