THE REDACTED ASSISTANT (2025–2026)

Most AI writing interfaces are designed to hand you a finished, smoothed-over surface — prose that arrives already closed, reducing the human writer to a passive editor of machine output. I built The Redacted Assistant to invert that dynamic: what if the model generated a dense, stylistically committed passage, ranked every word by its semantic weight, and then blacked out the load-bearing terms before you could get comfortable with them? Instead of a finished draft, the model gives you a haunted scaffolding of deliberate gaps that you have to inhabit, reconstruct, or write against.

You start with a prompt (or an uploaded .txt or .md file), one of five distinct writing registers, and a redaction density slider. In a single structured call, the model generates the passage and returns it as a JSON array in which every word and punctuation token carries a redaction rank: -1 for structural tokens that remain visible at all times, and positive integers ordering the remaining words by semantic importance, where lower numbers mark the most load-bearing concepts.

Because ranking happens once at generation time, the redaction density slider works entirely client-side: moving it filters to ranked tokens, computes how many words to conceal at the chosen percentage, finds the threshold rank at that cutoff, and masks everything at or below it instantaneously without another API call. From there, you can click individual black bars to reveal the underlying word, type your own replacements directly into the gaps, or export the result as a fill-in worksheet or a complete answer key.

The five built-in registers are not generic tone presets; each carries a system instruction anchored to a specific intellectual lineage — Hegel, Levinas, and Schelling for the philosophical mode; the CCRU’s theory-fictions for the hyperstitious mode. In production, generation calls proxy through a rate-limited Netlify function with sessions persisted in browser localStorage, and a Tauri wrapper packages the same frontend as a native macOS desktop tool.