LOGOI

AI Disclosure

AI Content Disclosure

The three kinds of authorship

Every Logoi surface falls into one of three authorship categories. The matrix below maps Logoi's surfaces to their authorship class.

Mechanical extract Audited primary corpus (passages, lemmas, morphology, lexicon entries)

Direct ingest of public-domain or compatibly-licensed editions (DIORISIS, Perseus, LSJ). Each passage carries provenance and an audit chain. No AI, no paraphrasing, no editorial reshaping at this layer.

AI-assisted Chat answers (/ and /ask, /api/v1/ask, MCP responses)

Logios, the librarian — a large language model configured as a restricted research librarian. It can only call Logoi's corpus tools; every answer streams with ai_authored: true, a disclosure line, source IDs, and a receipt URL. Words absent from the corpus return 'not attested' rather than an invention.

AI-assisted Etymology articles (/word/<slug>)

AI-drafted from parsed dictionary entries (Beekes, Chantraine, Frisk, LSJ) and live corpus statistics, with machine-verified citations against the parsed sources, reviewed before publication. Each article carries its own disclosure line.

Human Methodology, citation policy, legal pages, this page

Authored and edited by Cody Peterson. The architecture decisions, scholarly framing, and policy posture are human work.

Human Acknowledgments

Hand-curated upstream credits.

Logios, the librarian — what it is, what it is not

The four facts below answer the load-bearing questions a scholar or auditor would ask about the AI layer.

Who is Logios?

Logios (Λόγιος, 'skilled in words' — Hermes' epithet as god of eloquence) is Logoi's librarian: a retrieval-restricted AI. It answers only by calling Logoi's eighteen corpus tools (lemma lookup, concordance, passage lookup, lexicon, semantic search, morphology, frequency, and so on) against the audited database. It is not a general chatbot.

What grounds its answers?

Tool results only. Every claim in an answer traces to source IDs and a receipt URL streamed alongside the text. The assistant is instructed never to supply Greek text, counts, or etymologies from model memory.

What are its limits?

It paraphrases; it does not reprint copyrighted material. It cites every source. It flags uncertainty and honest absence ('not attested in the indexed corpus') rather than guessing.

Which model runs it?

Logoi's runtime is provider-agnostic: the assistant layer can run on different large language models behind a gateway with spend caps. The model in use is named in the meta event of every streamed answer.

Copyright posture

Three copyright positions, one per authorship category. The US Copyright Office's 2023 guidance on AI-generated content is the framing used here.

Mechanical extracts

Public domain or upstream-licensed (CC BY, CC BY-SA, etc.). Logoi's display does not change the upstream copyright. Attribution per source MANIFEST.

Human-authored prose (this page, methodology, etc.)

© 2026 Cody Peterson. All rights reserved. Reproduction with attribution permitted under CC BY 4.0.

AI-assisted prose (chat answers, etymology articles)

Per US Copyright Office 2023 guidance: human-curated/edited AI output is © Cody Peterson under the editorial-input doctrine. Purely-AI-generated text without human curation is public-domain in the US. Logoi's AI-assisted prose is human-curated by default; treat as © Cody Peterson unless otherwise noted.