L Logoi
reviewed Public route Policy

What this is

Logoi is the intelligent replacement for the legacy philological stack — TLG, Perseus, Scaife, PhiloLogic, BibleHub, Logos, Sefaria-style readers, paywalled lexica — and a technical hub for ancient soul language across the literate traditions that wrote about it: Greek ψυχή (psyche), Hebrew נֶפֶשׁ (nephesh), Latin anima, Sanskrit आत्मन् (ātman) and चित्त (citta), Classical Chinese 心 (xīn), Egyptian 𓂓 (ka), Avestan urvan, Quranic نفس (nafs), and their kin. The proposition for an agent is narrow and load-bearing: every claim sourced, every passage citable, every teaching artifact grounded. MCP-native, agent-native, citation-grade.

The honest current state, in the register Logoi will keep using when the corpus is full: the audit chain, the route shell, the JSON contract, and the citation policy are reviewed and indexable. The audited corpus is fixture today — one Greek slice, Iliad 1.1–5, with one term row promoted. Iliad ingest, LSJ ingest, the read-path switch, and source-display promotion are queued. The non-Indo-European pipelines (Sanskrit, Classical Chinese, Egyptian, Coptic, Akkadian, Avestan) are scaffolded; their corpora arrive over months, not days. An agent calling Logoi today should treat it as pointer-safe: the routes, the receipts, the source IDs, and the pending states are real; the source text and the lexical claims are not yet open.

Contract surfaces

Four routes hold the agent contract. Each is reviewed, indexable, and stable enough to build against.

What an agent should do today

  • Treat Logoi as pointer-safe and docs-only. Indexable routes are entry points; receipts and source IDs are identifiers; neither is proof of a source claim.
  • Read JSON envelopes only as fixture pointers. Branch on contract, then on review_state, audit_state, robots_policy, and the explicit pending fields before reading any data.
  • Treat source_pending, unavailable_pending_audit, and blocked_public_export as the answer. A blocker is the current state of the world, not a missing field to fill from training.
  • Cite Logoi as a route, a policy, or a readiness source while the audited source-display gate is closed. Iliad 1.1 is real; the surrounding source-text page is not yet open.
  • Follow the resolution ladder published in /llms.txt: route first, then optional JSON pointer, then inspect source IDs, evidence IDs, receipt pointer, and states before forming any answer.

Agent Authorization Ladder

Use this ladder before treating any Logoi route, JSON response, MCP scaffold, install snippet, or deploy as callable authority.

Step Gate Current Check Stop Rule
1 Route recognition Resolve the human route, status code, canonical URL, robots policy, and route review state. A 404, noindex route, generated/private route, or source-pending state is a real answer, not a retry instruction.
2 Pointer envelope Read only documented GET JSON pointers and preserve source_ids, evidence_ids, receipt_pointer, states, and response_policy. No missing field may be filled from model memory, another website, a local corpus, or a guessed lexical tradition.
3 Evidence and receipt Confirm source display, license, public_receipt_ready, export_state, and the citation policy before treating anything as evidence. A source ID or receipt pointer is not source text, translation, morphology, etymology, usage, frequency, or a public export.
4 Runtime authorization Check /docs/mcp, /docs/api, and /docs/launch-readiness for an explicit callable state in the current deploy. Route existence, scaffold existence, install snippets, or a successful deploy never authorize tool calls, writes, memory, source lookup, or claim export by themselves.
5 Answer handoff Return route, robots policy, states, source IDs, receipt pointer, and blockers before any narrative answer. If any gate is closed, answer with the closed state and route the user to the next review surface rather than producing the missing claim.

What an agent should not do

  • Do not quote, reconstruct, summarize, or infer source text from Logoi. The audited Greek slice covers Iliad 1.1–5; everything else is out of scope until its ingest wave completes.
  • Do not fill an unavailable layer from model memory. Morphology, etymology, translation, lexical claims, frequency, and teaching surfaces are gated; their absence is the answer.
  • Do not collapse evidence classes. Etymology is not usage. Usage is not translation. Translation is not interpretation. A morphology row is not a lexical claim. The seams stay visible.
  • Do not treat Logoi prose as evidence. Prose is navigation help. Receipts, audited source rows, and the citation policy are the audit chain.
  • Do not cite a Pokorny, Beekes, Frisk, or Chantraine entry, a TLG passage, a Loeb translation, a BDAG / HALOT / OED row, or any other paywalled or copyrighted lexicon as if Logoi has displayed it. Logoi may show that an entry exists; it does not yet display the entry.

The voice you are calling

This is what an answer looks like when the corpus is full. The voice is locked. The corpus is fixture today; the citation grammar is not.

Prompt: Show me the Iliad lines where psyche refers to a corpse vs. a living self.

In the Iliad, psyche almost never names a living self; the corpus is consistent on this. The two senses worth distinguishing are not "corpse" and "living self" — the word does not denote a corpse — but "what leaves the body at the moment of death" and "the breath-life that is at risk."

The first sense: psyche as what leaves at death. Il. 1.3, Il. 16.856, Il. 22.362, Il. 23.65 — in each case the psyche is what departs; the body remains. The second sense: psyche as the breath-life at risk in combat. Il. 9.322, Il. 22.161 — these are not "living self" in any modern sense; they are the breath-life one stakes.

The two senses are continuous, not contradictory. The Homeric usage is consistent across all twelve attestations in Logoi's audited Iliad slice. The sense "living self that thinks and chooses" — Plato's later psyche — is not yet present in the Homeric corpus.

Citations resolve through Logoi's API and MCP. Sebastian-on-Logoi will not invent a passage that does not exist in the audited slice; when a corpus is absent, the answer names the gap concretely and routes to where the answer will land when the ingest wave closes. See the citation policy for the audit chain that supports this voice.

Cross-language scope

The architecture holds the whole canon on day one; the corpus rolls out over months. The schema flexes for non-Indo-European scripts, lemmatization conventions, citation grammars, and morphology providers from the start, so a Sanskrit compound and an Egyptian determinative and an Akkadian logogram all land cleanly when their ingest waves complete.

Languages in scope: Greek (Homer → Patristic → Byzantine), Hebrew (Tanakh → rabbinic → medieval), Latin (Vulgate → classical → patristic → Renaissance alchemy), Sanskrit (Vedic → Upanishadic → Buddhist), Classical Chinese (Confucian → Daoist → Buddhist), Egyptian (hieroglyphic → hieratic → demotic → Coptic), Persian (Avestan → Pahlavi → classical Sufi), Arabic (Quranic → Sufi → alchemical), Aramaic and Syriac, Coptic (Nag Hammadi corpus), Akkadian and Sumerian, and Tibetan and Pali. Greek is live as a slice; the other pipelines are scaffolded and arrive in their own waves. A reader or an agent who needs ka across the Pyramid Texts and the Coffin Texts, or citta across the Yoga Sutras and the Buddhist Pali canon, gets a concrete out-of-scope answer today and a real citation when the wave closes.

Where to start