What the Lab released and why: retrieval preparation, document conversion, vector tooling, observability, learning programs, and the operating layer of our own cluster. Every entry is a public repository you can clone today, and every summary matches its README.
OSS/01..09Every release started as a tool the Lab needed: preparing documents for retrieval, converting formats, inspecting vectors, watching a fleet, printing a course, gating prose, operating two Sparks.
Register summaries repeat what each repository states about itself, nothing more. Where a number appears, the repository carries the measurement artifact behind it.
A tool can appear in two registers: yt2textbook is also Fieldwork FW-05, SparkDuet is also Local AI LA/01. Each has exactly one canonical detail page, and both registers point to it.
MIT and Apache-2.0 across the shelf. Clone, fork, ship. If a release saves your team a week, that is the intended outcome; if you need the discipline behind it, that is an engagement.
What the Lab open sourced and why: retrieval preparation, document conversion, vector tooling, observability, learning programs, and the operating layer of our own cluster. Every entry is a public repository you can clone today.
Summaries match each repository's README. Licenses and install paths live on the register page.
Document intelligence that prepares enterprise knowledge for retrieval, local-first: a browser-based pipeline that turns mixed repositories of PDF, DOCX, XLSX, EPUB, HTML, and text into clean, searchable chunks and JSONL records ready for RAG, with every byte processed client-side.
Nine public releases: DocPrep-AI (retrieval preparation), MarkForge (document conversion), Weaviate Local UI (vector tooling), DevOps Monitor (observability), yt2textbook (YouTube courses printed as textbooks), OneRead (a compiler for technical English), SparkDuet (the two-DGX-Spark operating layer), the AI Engineering Lab course, and AlchemyLake for Databricks. All under github.com/zorost.
The Lab publishes the tools it built for its own work when they are useful beyond it. Open source is the proof habit applied to tooling: the README is the claim, the repository is the evidence, and anyone can verify both.
MIT for most releases (MarkForge, OneRead, SparkDuet, AI Engineering Lab) and Apache-2.0 where noted (AlchemyLake for Databricks). Each repository states its license; the register entry mirrors it.
Yes. yt2textbook prints any YouTube video or playlist as an illustrated textbook with chapters, screenshots, comprehension checks, a glossary, and a Word export. It runs on your machine with your own model key; Zorost does not host a public runner.
OneRead type-checks technical English the way a compiler type-checks code: it parses prose, applies mechanical rules derived from ASD-STE100 Issue 9, and emits the result to ten surfaces including SARIF for CI, OpenAPI text, SKILL.md, and MCP tool descriptions.
We don't pitch slide decks. We show you what we've already built in your domain, then engineer what your mission requires.
