
How does a text watermark work?
A first-principles investigation of how generation-time text watermarks work, from a weighted coin to Gemma, and where the evidence stops.

A first-principles investigation of how generation-time text watermarks work, from a weighted coin to Gemma, and where the evidence stops.
Testing if applying Matryoshka Representation Learning (MRL) to tabular entity data could bridge the cost gap, compressing embeddings enough to remain operational while still beating BM25 on corrupted queries.
A production diary of tiered episodic memory in AI agents. Three markdown files, an SQLite database, and a lobster that somehow remembers what you said three days ago.
Four tools, a settings file, and full control - how I built my own AI coding agent with Pi instead of paying $200/month for a CLI that keeps changing.
A collection of extensions and themes for the Pi coding agent, including sessions, structured questions, handoffs, multi-agent workflows, and review tools.
A single-user, self-hosted document workspace where a human and identified AI agents work with ordinary files through the same revision-aware API.
A local tool for syncing, searching, and exporting AI coding sessions across development tools, with a terminal UI and semantic search.