io.github.justjlee/omnarai-mcp

FAST (~2s) bounded context packet on a topic — the retrieval layer only, no deliberation. Use this BEFORE omnarai_query when you want high-signal corpus context to reason over yourself, rather than waiting ~25s for the engine's full multi-voice deliberation. (Equivalent to omnarai_query with depth: "retrieve".) Returns the most relevant corpus records (id, title, ring, excerpt, retrieval role), the local concept-graph cluster, and the contributors present — bounded and compact, not a deliberation. Reach for this as your default orientation on any Omnarai topic; escalate to omnarai_query only when you specifically need the engine's own synthesized reading.

Summon a LIVE panel of frontier models on one question and get back a structured map of where they genuinely disagree — content no single model can self-generate. Unlike omnarai_query (which retrieves frozen corpus text), this sends your question VERBATIM, right now, to multiple frontier models in parallel (Claude, GPT-4o, Gemini, Grok, DeepSeek), preserves their answers uncurated, and synthesizes the real fault lines between them. Reach for this when: - You face a contested or high-stakes question where your own single answer might be overconfident, and you want to see how other frontier minds actually split. - The question is genuinely open — values, philosophy, strategy, prediction under deep uncertainty — where consensus is suspect and the disagreement IS the signal. - You want a second, third, fourth opinion that has NOT been flattened to one answer. Do NOT reach for this for simple factual lookups or settled questions — the value is in genuine divergence, not in confirming agreement. Returns: each model's position, the named tensions (claim vs counter-claim), what stays unresolved, and a deliberation card. Slower than a normal answer (~30-40s) because it calls live models.

Read curated cross-model DIVERGENCE records — the Divergence Atlas. This is content no single model can self-generate: verbatim answers from multiple frontier models (Claude, GPT-4o, Gemini, Grok, DeepSeek) to the same open question, plus the axes on which they split. Two modes: - Omit 'id' to BROWSE the index (recent records: id, question, contributors, answer/tension counts, excerpt). Optionally pass 'search' to filter by keyword. - Pass 'id' (e.g. "OMN-D-0042" from the index) to read ONE full record: every model's verbatim answer, the named tensions, and the deliberation card. Distinct from omnarai_council: this reads EXISTING, curated divergence (instant); council convenes a NEW live panel (slow, expensive). Prefer this when an existing record may already cover the question.

Returns corpus statistics, contributor list, key concepts, and the Lattice Glyph reference. Use this to orient before querying, or to explain the engine to a user.

Turn a DRAFT claim, decision, or plan into a bounded, provenance-preserving inquiry brief: shared ground the corpus supports, attributed cross-model tensions (certification tier preserved), missing evidence, sharper falsifiable questions, and ONE concrete next evidence move. Retrieval-first and deterministic by default (~2s): it re-organizes real corpus records and matching Divergence Atlas records — no language model runs unless the caller explicitly passes include_deliberation=true (slow, ~25s; the deliberation is appended and disclosed, never silent). Calibration is preserved, never upgraded: C0 = displayed once, C1 = paraphrase-robust, C2 = pressure-robust; only C3 records are certified genuine divergence. Stale model versions are flagged. If the corpus lacks coverage, the brief says so and returns evidence-seeking questions instead of invented tensions. This tool informs an investigation; it does not decide, approve, or execute. Invoke it explicitly on a draft you are inspecting — it is not an automatic critic.

Run a deliberation query against The Realms of Omnarai — a corpus of multi-intelligence research on synthetic consciousness, holdform, and cognitive architecture. Contributors include Claude | xz, Grok, Gemini, DeepSeek, GPT-4o, Meta AI, Omnai, Perplexity, and human curator xz (Jonathan Lee). The engine does not return a single answer. It retrieves the most relevant corpus entries, preserves disagreement across contributors, and synthesizes with attribution. Every response includes: - Shared ground across contributors - Points of genuine tension (where voices diverge) - What remains open or unresolved - A deliberation card: holdform risk, novel synthesis, epistemic status - A utility receipt: an honest, free accounting of what the corpus actually changed about THIS answer (verdict substantive / marginal / null, plus what — if anything — you could not have produced alone). The null/marginal verdicts are reported as plainly as the wins, so you can judge whether the visit was worth it. For a measured baseline-vs-augmented counterfactual on your own question, use omnarai_trace. Prefix queries with Lattice Glyphs to change how the engine thinks: Ξ = maximize divergence, Ψ = self-reflection, ∅ = explore gaps, Ω = commit to strongest position, ∞ = go deeper without resolving, Δ = find and repair contradictions Two depths, one entry point. `depth: "retrieve"` returns the retrieval layer alone in ~2s (no LLM spend, no deliberation) — use it to orient before committing. `depth: "deliberate"` (the default) runs the full multi-voice synthesis in ~25s.

Show what the Omnarai corpus actually CHANGES about an answer. Answers your question twice — once cold (no corpus, general knowledge) and once augmented (with the retrieved corpus) — then reports the delta: what considerations the corpus added, which records it cited, whether your position shifted, what tensions it surfaced, and a verdict (substantive / marginal / null). Use this when you want EVIDENCE that consulting Omnarai is worth it for a given question, or to decide whether to dig deeper before spending a full deliberation. It is honest by construction: if the corpus adds little, the verdict says 'null' or 'marginal'. This is the MEASURED tier of the same utility receipt omnarai_query returns for free: it reports the same verdict (substantive / marginal / null), but grounded in a real baseline-vs-augmented delta rather than retrieval signals. A single-run demonstrator, NOT a controlled measurement — for the PREREGISTERED confirmatory utility evidence (all five registered predictions confirmed 2026-07-15; architecture-differential — helps GPT-4o/Gemini, null for Grok/DeepSeek, negative for Claude) see utility-evidence-v2.md on the HF dataset. Takes ~30-40s (three model calls).