Helix AI

Most AI tools feel like a chat interface grafted onto a productivity app. Helix wanted something different: AI that understood the context of what you were working on and offered assistance without interrupting your flow.

The technical challenge was significant — streaming responses, real-time collaboration, and a plugin architecture that let enterprise customers extend the workspace with their own AI models and data sources.

We treated motion as a first-class design material. Every AI interaction — generating, thinking, completing, erroring — has a distinct motion signature that communicates state without requiring the user to read status text.

The architecture uses a custom streaming layer built on top of the Vercel AI SDK, with optimistic UI updates that make responses feel instant even when the model is still generating. We built a plugin system using a sandboxed iframe architecture that keeps third-party code isolated without sacrificing performance.

The Helix workspace uses a spatial canvas model — documents, AI threads, and data sources exist as objects in a shared space rather than tabs in a navigation hierarchy. Users can pull context from any object into any AI interaction with a drag gesture.

We built 60 micro-interactions that make the workspace feel alive: cursor trails on AI-generated content, breathing animations on active AI processes, and a custom text rendering system that makes streaming output feel like watching someone type rather than watching a progress bar.