Truth You Can See. Make AI outputs verifiable, auditable, and regulator‑ready.

Crebiliti transforms every AI statement into a measurable truth signal — detecting consensus, contradiction, and confidence in real time, and steering models toward verified trajectories.

Open interactive → Learn how it works

Built for enterprises that need certainty — not speculation.

V0/0No matches in corpus (Hallucination ?)

V241/10Near‑certain consensus

V0/65Strong contradiction

V71/88High ambiguity

V‑Graph Snapshot

live

V105 V88 V61 V24 V96 V90 V58 V97 V18 V84

Consensus Contradiction Open interactive →

How Crebiliti Works (at a glance)

AI Model Output

Raw sentences from any LLM.

AFU Engine

Extracts Atomic Fact Units (S–P–O + time, qualifiers, provenance).

V‑Graph

Consensus / contradiction lattice → V‑Scores with temporal & independence weighting.

Anchoring Layer

Adjusts token logits using V‑Scores; optional hard masks for high‑risk contradictions.

Signed Output

Enclave‑signed AR‑Bundle + V‑Badges (e.g., V95/5) for user & auditor.

AI Model Output

Raw sentences from any LLM.

AFU Engine

Extracts Atomic Fact Units.

V‑Graph

Consensus/contradiction → V‑Score.

Anchoring Layer

Logit shaping with V‑Scores.

Signed Output

Enclave‑signed AR‑Bundle + V‑Badges.

Why Enterprise would choose Crebiliti

Verifiable outputs — per‑sentence V‑badges with clickthrough evidence and timestamps.
Regulator‑ready — enclave‑signed audit bundles, reproducible trails.
Plug‑in minutes — no retrain required; compatible with major LLMs.
Continuous improvement — veracity‑regularized fine‑tuning reduces hallucinations over time.
Enterprise class Auditing — Clickthrough to verified statements in Corpus.

Explore APIs Talk to us

Example

“MINI was first produced in 1959.” → V95/5

“MINI debuted in 1940.” → V12/88 (contradicted)

Security

ARM TrustZone / NVIDIA Grace CCA attestation with code hash, model hash, V‑Graph snapshot, and monotonic counters.

Performance

Sub‑200ms V‑lookup with warm cache; graceful fallback to uncertainty surfacing in sparse domains.

APIs

POST /score

Text → AFUs + V‑badges + evidence bundle.

POST /anchor

Logits + AFU context → adjusted logits + audit token.

POST /train-hook

Targets → veracity‑regularizer weights and manifold pulls.