Niches · Guardrails & safety filters

Check a RAG answer is supported by the retrieved sources

Two or three companies' docs show a full fit.

2companies, full fit
5companies, fit with a limit
15companies carry the tag
5share of the category's work
3first-hand accounts

What counts as done

The check takes the sources and the answer as separate fields and returns a grounding or relevance score per answer with a documented threshold, and an ungrounded answer can be blocked or flagged before it is shown.

How agents ask for it: is this answer grounded in the documents, flag hallucinations in our RAG bot, check the answer stays on topic

Does the job 2

The docs show it meets the test above, and nothing we read says otherwise.

ListingWhy
Amazon Bedrock Guardrails A
Amazon Web Services, Inc.
Contextual grounding checks take the grounding source, query and response as separate tagged inputs and return grounding and relevance scores against thresholds you set, up to 100,000, 1,000 and 5,000 characters. source
NVIDIA NeMo Guardrails BB
NVIDIA Corporation
Fact-checking rails take relevant_chunks as evidence and the bot message separately, AlignScore (a RoBERTa model, no LLM call) or self check facts return a score, blocked below 0.5 in the documented flow. source

Does the job, with a limit 5

It does the job, but something in the docs or a user's account will get in an agent's way. The limit is named in each row.

ListingWhy
Azure AI Content Safety (Prompt Shields) B
Microsoft Corporation
Groundedness detection takes grounding sources (up to 55,000 characters) and the answer separately, with reasoning and correction modes, but it is preview and retires on 30 April 2027, English only. source
Granite Guardian B
International Business Machines Corporation
Groundedness, context relevance and answer relevance take documents and the answer separately, but each is an 8B-model call per criterion returning yes or no, the extra-call blocker. source
Guardrails AI D
Guardrails AI, Inc.
Provenance Embeddings, LLM and NLI validators check output against sources per the hub catalog, but each needs a local model or an extra LLM call and no threshold docs were read. source
OpenAI Guardrails BB
OpenAI (as named in the LICENSE copyright line and the PyPI author field)
Hallucination detection checks answers against an OpenAI vector store you create, with a threshold, but it is an extra LLM call per turn and tied to OpenAI file search. source
Prisma AIRS AI Runtime Security API C
Palo Alto Networks, Inc.
A context field takes the grounding data separately and the response reports ungrounded true or false, but no score or threshold is documented. source

Doesn't do this job 10

Listings in the category that do something else, with the reason.

ListingWhy
AgentGate D
Mohit Singh (individual)
No grounding check. source
Cisco AI Defense Inspection API B
Cisco Systems, Inc.
No grounding or relevance rule in the docs or SDK read. source
Google Cloud Model Armor BB
Google LLC
No grounding or relevance check; filters cover safety, injection, sensitive data, malicious URLs and CSAM. source
Lakera Guard (Check Point AI Guardrails) B
Check Point Software Technologies Ltd.
No grounding or relevance detector among the guardrails listed; detectors cover prompt attacks, moderation, data leakage, links and agent behaviour. source
Llama Guard 4 D
Meta Platforms, Inc.
No grounding check. source
LlamaFirewall D
Meta Platforms, Inc.
No grounding scanner. source
Presidio B
Data Privacy Stack (community organisation, no legal entity stated)
No grounding or relevance check. source
Mistral Moderation API B
Mistral AI (RCS Paris 952 418 325)
No grounding check among the categories. source
OpenAI Moderation API BB
OpenAI OpCo, LLC
No grounding check. source
ShieldGemma C
Google LLC (RDAP registrant of gemma.ai; the Gemma Terms of Use are Google's and the cards name Google as author)
No grounding check. source

What stops agents finishing

  • grounding check needs an extra LLM call per turn
  • check applies only inside the vendor's own model invocation

Around it

Before: vector-search.query, search.ground-fact. After: human-in-the-loop.review-output.

Where this comes from

Weight 5 of 100 in Guardrails & safety filters: 3 of 50 accounts (one is a recommendation, not a report of use); Bedrock, Azure, NeMo and Granite Guardian all advertise it, so it is kept at a low weight

The same as data: /api/v1/use-cases/guardrails.grounded-answers.json, or the MCP tool get_use_case with id guardrails.grounded-answers. Other jobs in this category: /api/v1/niches?category=guardrails.

Draft, read 2026-11-15 from public docs and first-hand accounts, not yet reviewed by a person. A mark says what the docs show, not what a run measured.

For companies

Do agents find, use and choose your tools?

An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.