Visual spaghetti for AI agents, no coding required
A node-based IDE that lets you wire together LLM pipelines without touching a terminal.

What it does
Magick is a visual development environment for building AI agents and data pipelines. You drag nodes onto a canvas, wire them together, and export the result as JSON “spells.” It handles prompts, code execution, image generation, vector search, and social connectors (Discord, Twitter, Twilio) in one graph-based interface.
The interesting bit
The project treats “spells” as first-class, portable artifacts—entire agent pipelines serialized to JSON that can be shared or embedded as subgraphs. That’s a bet on composability over lock-in, though the README is upfront that “spell is not a machine learning term. We just like it.”
Key highlights
- Realtime multimodal agents with unified memory that can act autonomously or interact with users
- Pre-built nodes for voice, image generation, vector search, and web search (Google, Wikipedia, Semantic Web)
- Social connectors to Discord, Twitter, Twilio; Zoom, Slack, Reddit listed as future plugins
- Subgraph embedding for nesting and community reuse of pipeline components
- Cross-platform with dedicated macOS, Linux, and Windows install guides
Caveats
- Several promised connectors (Zoom, Google Meet, Reddit, Slack) are listed as “available soon as plugins”—not shipped yet
- Self-signed certificate workarounds required for local development, suggesting rough edges in the dev setup
- 837 stars with 27 contributors; active but not battle-tested at scale
Verdict
Worth a look if you’re building chatbots or agent prototypes and prefer visual tools to YAML wrestling. Skip it if you need production-grade reliability or already live comfortably in code.