Antispace AI is a unified AI-native productivity OS. It combines all of your productivity tools and third party services in a single AI-driven suite.
Integration System
At Antispace, I built a complete platform that lets developers write custom integrations for the platform.
The system has 3 main parts: a TypeScript SDK, a CLI tool for development and deployment, and a server-side renderer that converts JSON to React components. Apps run in isolated environments with no client-side code and no access to user data - everything renders server-side for security.
CLI Tool
I wrote a CLI (`@antispace/cli`) that handles the full development workflow. Run `anti dev` and it spins up a local server with hot reloading, watching your source files and regenerating route handlers on the fly. When you're ready to ship, `anti build` bundles everything with Bun, analyzes your TypeScript to wire up endpoints automatically, and creates a single deployable build artifact.
The build process parses action functions, manifest definitions, and UI components to generate a single entry point that handles routing, AI function calls, and knowledge map updates.
The build process parses action functions, manifest definitions, and UI components to generate a single entry point that handles routing, AI function calls, and knowledge map updates.
Rendering the UI
I also created a UI system where developers can write AI-friendly JSON component structures that render as system-defined React elements. The SDK exports components like `Column`, `Button`, `Input`, etc, with TypeScript props. The frontend gets this JSON and renders it with a React-based interpreter.
That included a `propTypes` feature that lets the AI fill in component properties dynamically based on function responses, instead of developer having to hardcode response strings, which makes integrations a lot more lively and dynamic in this AI-driven system.
Platform Services
Early versions of this system allowed deploying user integrations straight to Antispace's distributed infrastructure.
I also built multiple DX improvement services like KV storage for app persistence, semantic embedding ingestion for the knowledge graph, and rate limiting tool for user's upsterams with per-function budgets.
The platform supports OpenAI function calling so users can trigger integrations through natural language.