# PrismShine

One canonical identity. Human page: https://www.insightits.com/products/prismshine.html

| Field | Value |
|-------|-------|
| Slug | `prismshine` |
| Status | published |
| Family | aiVerification |
| Infra category | AI Verification |
| Record type | product |
| Parent | — |
| JSON | https://www.insightits.com/catalog/prismshine.json |
| Markdown | https://www.insightits.com/catalog/prismshine.md |
| GitHub | https://github.com/insightitsGit/PrismShine |
| PyPI | https://pypi.org/project/prismshine/0.2.2/ |
| Install | `pip install prismshine (pin ==0.2.2 in docs). Apache-2.0 open core. Pro/Enterprise are pre-validation estimates.` |
| Version | 0.2.2 |

## What it is

Self-hosted anti-hallucination verdict engine: cause-side forensics (Tier-0 halt before tokens) plus effect-side grounding (Tiers 1–4) in one auditable ShineVerdict. PASS means grounded in the supplied preload — not world-true.

## Problem

Fluent answers are treated as truth. Teams need an enforceable pass / block / review gate plus evidence hashes, not only an offline faithfulness score.

## Alternatives

- **HHEM-2.1-Open (Vectara)** — Published vendor-authored HaluEval comparison on identical data and Azure ACI hardware. HHEM is a strong open factual-consistency classifier; PrismShine is an evidence-aware verdict engine with a runtime halt path. ([source](https://www.insightits.com/compare/prismshine-vs-hhem.html))

## Features

- **Cause + Effect Gate** — Tier-0 halts empty retrieval, tool errors, and stale cache before tokens. Tiers 1–4 ground answers after generation.
- **Auditable ShineVerdict** — Named resolution_gate + evidence_hash on every decision — not a black-box score.
- **Zero LLM on Default Path** — Optional Tier-4 judge only on gray zone. pip install prismshine — no license key for OSS.

## Architecture

Tier 0 can halt empty retrieval, tool failure, stale cache, or incomplete trace before generation. Tiers 1–4 check answer support against the supplied preload. Optional ONNX Tier 3 (~1 GB, not in the bare wheel). ChorusGraph extra: pip install "prismshine[chorusgraph]" then require_shine / shine_node.

## Use cases

- Gate answers before they leave the agent
- Cause-side halt when retrieval or tools fail
- Auditable ShineVerdict with evidence hashes

## Benchmarks

Vendor-authored Azure ACI receipt 2026-07-20_run4_onnx: HaluEval QA n=200 F1 0.831 vs HHEM 0.746; numbers n=50 F1 1 vs 0.926. PASS means grounded in the supplied preload, not world-true.

### Disclosures

- The benchmark is vendor-authored.
- Bsum absolute F1 is modest for both systems.
- Tier-3 ONNX is an optional approximately 1 GB artifact and is not included in the bare wheel.
- The default path is English-centric.
- No RAGAS performance row has been completed.
- Do not present the evidence-aware runtime track as like-for-like content accuracy.
- No claim against RAGAS faithfulness.
- No claim against Guardrails AI provenance validators.
- No claim against NVIDIA NeMo Guardrails.

### Evidence URLs

- https://github.com/insightitsGit/PrismShine/blob/main/docs/BENCHMARKS.md
- https://github.com/insightitsGit/PrismShine/tree/main/benchmarks/progress/2026-07-20_run4_onnx
- https://github.com/insightitsGit/PrismShine/blob/main/benchmarks/progress/2026-07-20_runtime_docker/FULL_REPORT.md


## GitHub

https://github.com/insightitsGit/PrismShine

## PyPI

https://pypi.org/project/prismshine/0.2.2/

## Documentation

- [Python hallucination detection guide](https://www.insightits.com/guides/hallucination-detection-python.html)
- [PrismShine vs HHEM](https://www.insightits.com/compare/prismshine-vs-hhem.html)
- [Hallucination detection tools](https://www.insightits.com/compare/hallucination-detection-tools.html)
- [Benchmarks](https://github.com/insightitsGit/PrismShine/blob/main/docs/BENCHMARKS.md)
- [Limits](https://github.com/insightitsGit/PrismShine/blob/main/docs/LIMITS.md)
- [Integration](https://github.com/insightitsGit/PrismShine/blob/main/docs/INTEGRATION.md)
- [https://github.com/insightitsGit/PrismShine/tree/main/benchmarks/progress/2026-07-20_run4_onnx](https://github.com/insightitsGit/PrismShine/tree/main/benchmarks/progress/2026-07-20_run4_onnx)
- [readme](https://github.com/insightitsGit/PrismShine/blob/main/README.md)
- [interactiveDemo](https://insightitsgit.github.io/PrismShine/demo.html)
- [receiptRuntime](https://github.com/insightitsGit/PrismShine/tree/main/benchmarks/progress/2026-07-20_runtime_docker)
- [fullReport](https://github.com/insightitsGit/PrismShine/blob/main/benchmarks/progress/2026-07-20_runtime_docker/FULL_REPORT.md)
- [positioning](https://github.com/insightitsGit/PrismShine/blob/main/docs/POSITIONING.md)
- [enterpriseWiringDemo](https://github.com/insightitsGit/PrismShine/blob/main/examples/enterprise_wiring_demo.py)

## Is not

- A firewall (that is PrismGuard)
- An agent runtime (that is ChorusGraph)
- A claim that PASS means world-true
- Streaming hallucination protection for already-streamed tokens
- A RAGAS performance row (not published)

## Related products

- [chorusgraph](./chorusgraph.md)
- [prismguard](./prismguard.md)
- [prismrag](./prismrag.md)