Response Architecture
Three Response Tiers. One Platform.
Axiom dynamically selects the optimal response tier for every query. All tiers are auditable, all answers are grounded in knowledge.
Tier 1 — Deterministic
Pure Knowledge Lookup
Direct knowledge-pack hit. No LLM involved. Zero hallucination, sub-2ms, cryptographically auditable. The gold standard.
- Zero-hallucination guarantee
- Sub-2ms response time
- 181,000+ auditable topics
- Guided choice disambiguation
- Enterprise compliance ready
Pure Lume — Zero LLM
Tier 2 — Grounded Composition
Fact-Constrained Generation
Retrieves the most relevant facts from the knowledge base, then composes a fluent response using only those facts. Auditable sources, natural language.
- Knowledge-retrieval-first
- LLM constrained to verified facts only
- Source citations in every response
- No parametric speculation allowed
- Fluent, natural language output
Fact-Grounded — Auditable Sources
Tier 3 — Conversational
Lume-V Governed LLM
Full conversational fluency governed by the Lume-V trust layer. Used only when no relevant knowledge exists. Always labeled as conversational.
- Lume-V governance wrapper
- Claude 3.5 Sonnet backend
- Multi-turn conversation memory
- Personality engine (Professional/Casual)
- Clearly labeled as non-deterministic
Lume-V Governed — Claude Fallback