Case study
Department of Curious Notions
A local-first AI idea workspace presented as a living, animated government office, with a structured docket system, AI-assisted development, and a full agent runtime under the hood.
Purpose
A local-first AI idea workspace presented as a living, animated government office, with a structured docket system, AI-assisted development, and a full agent runtime under the hood.
What this is
The Department of Curious Notions is a personal AI idea workspace built to look and feel like a small animated government office, with a visible cast of workers, a docket system for developing ideas, a Director's Brief, and an After-Hours mode for open-ended exploration.
The framing is deliberately silly. The architecture under it is not.
What's under the hood
The app runs as a Vite and React frontend with a Node API behind it. The AI layer has three provider modes. There is a deterministic mock for development and offline work, a local Ollama path that runs Apple Silicon models on the Metal GPU, and the hosted Anthropic API. Every run logs a provider trace that shows which model ran, whether it fell back to another, and why. Nothing happens silently.
Data lives in SQLite and exports cleanly to JSON. The whole stack runs in Docker behind a single bearer token, and a backup is one npm run backup away.
When the system thinks a new capability is needed, it raises a tool-gap proposal, and nothing gets scaffolded or written to the workspace until I approve it. The idea is that the AI proposes and the human decides.
Why this is in the portfolio
The workspace is personal, but the way it is built is the point. It is local-first, inspectable, provider-agnostic, and gated with explicit safety checks and traceable logs. Those are the same things that make an AI tool trustworthy inside an organization, which is what makes a whimsical side project relevant here.
Related