Alfred
I operationalised my cognition into a tool for writers, solving the problem - how to turn high-dimensional thinking into legible artifacts without losing the author's voice.
Core trick
Restricts the model to four structural operations (split, merge, move, hoist) so it can reorder text but cannot generate new prose.
Tags
Description
I operationalised my cognition into a tool for writers, solving the problem - how to turn high-dimensional thinking into legible artifacts without losing the author's voice.
You spend 30% of your writing time just... **moving things around**.
Not writing new ideas. Not researching. Just reorganizing. You have three paragraphs saying the same thing. You have one paragraph doing too much. You have ideas in the wrong section.
You know what you want to say, but untangling it, so that others understand exactly what you mean, is a very time taking task.
All AIs that try to solve this whitewash your voice.
Meet Alfred, it is the Photoshop for writing structure. You write messy. It photoshops it into coherence. **It doesn't rewrite in AI voice. It refactors your structure while keeping your words.**
Alfred constrains AI to operate only on structure — Split, Merge, Move, Hoist — and shows you in real time what it's learning about how you think. The thesis is that architectural constraint, not prompt instruction, is what preserves authorship when AI enters the loop.
README excerpt
# Alfred Alfred is a writing environment where AI can rearrange your writing, but not overwrite your voice. Alfred is the Anthropic Opus 4.7 hackathon submission for the **"Build for What's Next"** track. I was selected as one of 500 builders, out of 20,000 applicants globally in Anthropic's Hackathon (2.4% acceptance), and I received 500$ of Claude API credits for the same. https://cerebralvalley.ai/e/built-with-4-7-hackathon/hackathon/gallery?project=233 ## What it does Most AI writing tools generate prose. They're great at average sentences. They flatten voice. The result is "whitewashing" — your writing returns to you smoother and duller. Alfred inverts that. Claude is given a **fixed operator algebra** — `split`, `merge`, `move`, `hoist`, `demote`, `migrate`, `glue` — and is architecturally forbidden from authoring free prose. It can move your text around. It can collapse redundancies. It can promote a buried thesis. It cannot rewrite your sentences. The result: a tool that helps with structure without touching voice. The author keeps every word. The architecture is enforced by a **Voice Guardian** that validates every proposal before showing it: glue text is capped at 15 tokens per operator (60 total), forbidden tokens are blocked, and the only operator that may rewrite words (`migrate`, used for reprojecting older fragments into your current voice frame) is capped at 50% token-edit distance. A side panel — the **Panopticon** — shows you, in real time, what Al