Question 1

What is new in PrismGuard 0.1.6?

Accepted Answer

Shadow ONNX (score the model without blocking), feedback persistence + export, and prismguard-model train / eval / calibrate so you can build a hub or customer ONNX artifact from your traffic. Rules-first web_chat remains the safe default until that artifact passes gates.

Question 2

Does PrismGuard require OpenAI?

Accepted Answer

No. The default classifier path uses a local ONNX model. OpenAI is only needed if you enable the optional LLM Judge on gray-zone traffic (Business tier).

Question 3

Should I turn ONNX on for general product / FAQ chat?

Accepted Answer

Not with the law proof artifact alone. prism-pi-v1 is calibrated for law-bench traffic and can false-positive greetings and pricing FAQs. Use PRISMGUARD_SHADOW_ONNX=1 + feedback, train/eval a hub or customer artifact, then set PRISMGUARD_USE_ONNX=1 with PRISMGUARD_ARTIFACT_ID or PRISMGUARD_GUARD_MODEL_PATH.

Question 4

Where does the ONNX model come from?

Accepted Answer

After pip install with the guard-model extra, run prismguard-model download (default: prism-pi-v1 law proof). Customer/hub weights come from prismguard-model train (or a published hub artifact when available) and are selected via ARTIFACT_ID / GUARD_MODEL_PATH.

Question 5

Is traffic sent to a third-party scanner SaaS?

Accepted Answer

No by default. PrismGuard is self-hosted — prompts stay on your infrastructure for the rules + ONNX path. That is the point for compliance buyers.

Question 6

What is resolution_gate?

Accepted Answer

Every allow/block decision exposes resolution_gate and decision_source so compliance can see which stage decided (rules, taxonomy/fusion, ONNX model, or optional Judge). With shadow ONNX, the model score appears in details.shadow_onnx without enforcing.

Question 7

Is PrismGuard only for legal / law firms?

Accepted Answer

No. PrismGuard is the ChorusGraph security plug-in for any production LLM stack — legal, healthcare, finance, and general copilots. The law pack and prism-pi-v1 cold-holdout are published proof, not a product limit. Align ONNX to your domain with shadow + feedback + train on your traffic (0.1.6+).

Question 8

How does PrismGuard plug into ChorusGraph?

Accepted Answer

Use make_guard_handler(checker) as the first graph node before cache, PrismRAG retrieval, and the LLM. Every decision exposes resolution_gate for compliance logs. Same pattern we dogfood on insightits.com Website and Dashboard hubs.

Question 9

Is healthcare or finance ready?

Accepted Answer

Domain packs for healthcare and finance are not claimed ready yet. The core rules firewall still applies; use shadow + feedback + train / custom lexicon / pilot for domain tuning. Do not treat holdout YAML as a customer runtime feature.

Question 10

How is this different from LLM Guard?

Accepted Answer

LLM Guard is a toolkit of scanners that return scores. PrismGuard is a firewall: allow/block policy with an auditable resolution_gate on every decision. They can be complementary; PrismGuard is built for production policy and compliance logs across domains.

Question 11

What is the Production Pilot vs OSS?

Accepted Answer

OSS (pip install) gives you the library, ONNX path, CLI, shadow/feedback train gears, and domain packs (law verified). The ~$25k Production Pilot (pre-validation estimate) adds staging deploy, custom lexicon, audited decision-log design, ChorusGraph or HTTP sidecar wiring, and 90-day Slack + SLA — with a conditional 14-day refund guarantee.

Question 12

Can non-Python stacks use PrismGuard?

Accepted Answer

Yes on the Business tier: prismguard serve exposes POST /v1/check, POST /v1/scan-output, and GET /metrics. Production serve requires a PRISMGUARD_LICENSE_FILE.