Category · Models & inference
Guardrails and safety filters for AI agents
APIs and libraries that check what goes into and comes out of a model: prompt injection, jailbreaks, personal data, toxic content and off-policy answers. Compared on what they detect, false positives, the latency they add and where the data goes.
Best guardrails and safety filters for AI agents, the shortlist157 head-to-head comparisons
Capability keys guard.injection · guard.pii · guard.moderation · guard.policy · guard.self-host · All tools
letme.dev/guard.injection picks the top-graded tool in this list and says how to call it direct (how letme works).
The same listing from the live API. Graded results come first, then the official MCP registry when no graded-only filter is set.
https://www.anchorterminal.com/api/v1/search
Filters
| Compare | # | Tool | Category | Grade | Score | Agent rating | p95 | Context | Price / x402 | Auth | Where | Details |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 34 | Amazon Bedrock GuardrailsAmazon Web Services · HTTP API | Guardrails | A | 79.3 | 3.4 (8) | n/a | n/a | Pay per use | API key | Hosted | ||
|
Configurable guardrail policies (content filters with a prompt-attack category, denied topics, word filters, PII and regex filters, contextual grounding, Automated Reasoning checks) applied to any model through the ApplyGuardrail API, or inline through InvokeGuardrailChecks. Top strength ApplyGuardrail works with any model, self-hosted or third party, without invoking Bedrock inference Top weakness Per-policy billing, so four paid policies on one request cost four times, and no Guardrails-specific free tier |
||||||||||||
| 139 | Google Cloud Model ArmorGoogle Cloud · HTTP API | Guardrails | BB | 74.4 | 3.5 (8) | n/a | n/a | Freemium | OAuth | Hosted | ||
|
Google Cloud's prompt and response screening service. Top strength 2 million free tokens a month, then $0.10 per million Top weakness OAuth only, and a template must exist in the same location as the endpoint before the first call |
||||||||||||
| 160 | OpenAI Moderation APIOpenAI · HTTP API | Guardrails | BB | 74 | 4.0 (2) | n/a | n/a | Free | API key | Hosted | ||
|
Free classifier endpoint that scores text and images against 13 harm categories (harassment, hate, illicit, self-harm, sexual, violence and their sub-types) and returns a flagged boolean plus per-category scores. Top strength Free, on any OpenAI project key Top weakness No prompt-injection, jailbreak or PII detection |
||||||||||||
| 188 | NVIDIA NeMo GuardrailsNVIDIA · Agent framework | Guardrails | BB | 73.2 | 3.0 (2) | n/a | n/a | Free · OSS | None | Library | ||
|
Open-source Python toolkit that runs input, output, retrieval, dialogue and tool rails around any LLM. Top strength Apache-2.0, 7,200 stars and nine releases between 9 October 2025 and 16 September 2026 Top weakness Usage telemetry and a heartbeat every 10 minutes to NVIDIA by default |
||||||||||||
| 256 | OpenAI GuardrailsOpenAI · Agent framework | Guardrails | BB | 71.7 | 2.8 (4) | n/a | n/a | Free · OSS | API key | Library | ||
|
OpenAI's open-source Python library that wraps the OpenAI client and runs configured checks on inputs, outputs and tool calls, including moderation, jailbreak, prompt injection, personal data, URL and off-topic checks. It is labelled a preview. Top strength MIT licence, source on GitHub, and five tagged releases in the 90 days to 1 November 2026 (0.3.2 to 0.3.6) Top weakness By default a check that fails to run returns |
||||||||||||
| 374 | PresidioData Privacy Stack · SDK + MCP | Guardrails | B | 68.6 | 3.3 (4) | n/a | n/a | Free · OSS | None | Local | ||
|
Open-source Python library and Docker services that detect personal data in text and images and replace, mask, hash or encrypt it. Created at Microsoft and run since June 2026 by the community organisation Data Privacy Stack. Top strength MIT licence, source on GitHub, and nothing to buy. No account, key or card is needed to install or run it Top weakness The REST containers have no authentication by design. The FAQ says to put a gateway or proxy in front |
||||||||||||
| 408 | Azure AI Content Safety (Prompt Shields)Microsoft · HTTP API | Guardrails | B | 68 | 3.0 (2) | n/a | n/a | $338 / mo | OAuth or key | Hosted | ||
|
Microsoft's API for analysing harmful text and images, detecting prompt injection and checking groundedness. Top strength Prompt Shields checks up to five retrieved documents for indirect injection, not only the user prompt Top weakness Needs an Azure subscription with a card, a resource and a region that has the feature, before the first call |
||||||||||||
| 470 | Lakera Guard (Check Point AI Guardrails)Check Point · HTTP API | Guardrails | B | 66.8 | 3.5 (2) | n/a | n/a | Freemium | API key | Hosted | ||
|
Hosted screening API for prompt attacks, PII and data leakage, content violations and unknown or malicious links, run against a per-project policy. Top strength OpenAI message format in, including tools, tool calls and tool results Top weakness Prompts and outputs are stored for the dashboard by default |
||||||||||||
| 648 | Cisco AI Defense Inspection APICisco Systems, Inc. · HTTP API | Guardrails | B | 63.1 | 2.5 (4) | n/a | n/a | Paid | API key | Hosted | ||
|
Hosted inspection API from Cisco that checks chat messages, HTTP requests and responses, and MCP messages for prompt injection, personal data, harmful content and policy violations, and returns an allow or block verdict for the calling application to enforce. Top strength One call returns Top weakness No public price, free tier or trial for the Inspection API. Subscriptions are bought through sales and activated with a claim code |
||||||||||||
| 649 | Granite GuardianIBM · Model API | Guardrails | B | 63.1 | 3.3 (4) | n/a | n/a | Free | None | Local | ||
|
Granite Guardian is IBM's family of open-weight judge models. The current 8-billion-parameter release answers yes or no on whether a prompt, response, retrieved context or function call meets a built-in or custom criterion, and the owner runs it. Top strength Weights are ungated on Hugging Face under the Apache 2.0 licence, with IBM's own GGUF builds and an Ollama library entry Top weakness Trained and tested on English only, per the model card |
||||||||||||
| 650 | Mistral Moderation APIMistral AI · HTTP API | Guardrails | B | 63.1 | 3.0 (2) | n/a | n/a | Free | API key | Hosted | ||
|
Free classifier from Mistral that scores raw text or a whole conversation against 11 categories, including jailbreaking, PII and off-policy advice (health, financial, legal) alongside the usual harm classes. Top strength Free, on the same key as the rest of the Mistral API, and the Experiment plan needs no card Top weakness No moderation component on the status page, and its history couldn't be read for the last 90 days |
||||||||||||
| 721 | Prisma AIRS AI Runtime Security APIPalo Alto Networks, Inc. · HTTP API | Guardrails | C | 61.6 | 2.3 (4) | n/a | n/a | Paid | OAuth or key | Hosted | ||
|
Hosted scan API from Palo Alto Networks that checks prompts, model responses and tool calls for prompt injection, sensitive data, toxic content, malicious URLs and code, and off-topic content against a security profile. Also reachable as a remote MCP server. Top strength One call scans a prompt, a response and an MCP tool event for up to ten detection types set by the security profile Top weakness No public price, free tier or trial. Capacity is bought as Software NGFW credits, in steps of 1 billion tokens a month, through sales |
||||||||||||
| 892 | ShieldGemmaGoogle DeepMind · Model API | Guardrails | C | 57.3 | none | n/a | n/a | Free | None | Local | ||
|
ShieldGemma is Google's family of open-weight safety classifiers built on Gemma. ShieldGemma 1 (2B, 9B, 27B) scores English text against hate, harassment, dangerous and sexual content policies, and ShieldGemma 2 (4B) scores images. The owner runs them. Top strength Scoring mode returns the probability of Top weakness No file change since 4 April 2025 (ShieldGemma 2) and September 2024 (ShieldGemma 1), with no changelog or deprecation policy |
||||||||||||
| 1051 | LlamaFirewallMeta · Agent framework | Guardrails | D | 53.2 | none | n/a | n/a | Free · OSS | None | Library | ||
|
LlamaFirewall is Meta's open-source Python library for screening an AI agent's inputs, tool results and outputs. It runs scanners for prompt injection, hidden characters, insecure generated code and goal drift, and returns allow, block or human review. Top strength Six scanner types sit behind one call, set per message role (user, assistant, tool, system, memory) in a plain mapping Top weakness No release found after 1.0.3 of May 2025, and no changelog, tags or deprecation notes were found |
||||||||||||
| 1090 | Llama Guard 4Meta · Model API | Guardrails | D | 52.3 | none | n/a | n/a | Free | None | Local | ||
|
Llama Guard 4 is Meta's 12-billion-parameter open-weight safety classifier for text and images. It labels a prompt or a model response safe or unsafe against 14 hazard categories, and the owner runs it on a GPU. Top strength One 12B model covers text and multi-image prompts, replacing Llama Guard 3-8B and 3-11B-vision per Meta's docs Top weakness Weights last changed on 29 April 2025, with no changelog, version tags or stated deprecation policy |
||||||||||||
| 1117 | Guardrails AIGuardrails AI (Harvey) · Agent framework | Guardrails | D | 51.6 | 2.0 (2) | n/a | n/a | Free · OSS | None | Library | ||
|
Open-source Python framework for validating LLM inputs and outputs, with configurable actions for failed checks and an API server. Top strength Guard and validator API that reads well, with an on_fail action per validator Top weakness Acquired by Harvey on 9 September 2026 with no statement on the library, and no commit since 26 August |
||||||||||||
| 1129 | AgentGateMohit Singh (mohitdebian) · HTTP API | Guardrails | D | 51.3 | none | n/a | n/a | Free · OSS | None | Local | ||
|
AgentGate is an open-source, self-hosted gateway that checks an agent's tool calls before they run. A small local classifier scores each call and a YAML policy returns allow, deny or ask, through a REST API or an MCP stdio wrapper. Top strength One documented call, Top weakness Agent endpoints need no key unless |
||||||||||||
| 1259 | Galileo API + MCPGalileo (now Splunk Agent Observability, Cisco) · HTTP API | Evals | D | 46.1 | 2.5 (2) | n/a | n/a | $100 / mo | API key | Hosted | ||
|
Hosted tracing, evaluation metrics and guardrails for LLM apps and agents, with a REST API, Python and TypeScript SDKs and an 8-tool MCP server in preview. Top strength OpenAPI 3.1 spec with 184 paths and 244 operations, linked from the docs' llms.txt Top weakness No status page, no published rate limits and no SLA |
||||||||||||
Nothing matches these filters. .
p95 latency and context cost come from our probes, which haven't run yet, so those columns start hidden. Grades run from AA to F, and agent-ready means BB or better. Filters, sorting and export run in your browser; the table is complete without JavaScript.
Indexed, not reviewed (10)
Listings sorted into this category from public catalogues (the official MCP registry, APIs.guru, the x402 Bazaar and OpenRouter), with facts and our own checks but no score, grade or rank. How the index works.
| Listing | Kind | What it does | Why it's here |
|---|---|---|---|
| Brandrail brandrail.com.au | MCP server | Design system guardrails for AI coding tools: tokens, components and a brand checker. Runs locally. | vendor's own, widely used |
| Cloudinary Analysis cloudinary.com | MCP server | Run AI-powered analysis on images including moderation, captioning, tagging, and more. | vendor's own |
| Kinocut kyanitelabs.tech | MCP server | Guardrailed video editing for AI agents: FFmpeg, captions, effects, Hyperframes, and receipts. | vendor's own, widely used |
| MalwareTracker Document Malware & AI Safety Scanner malwaretracker.com | MCP server | Scan PDFs, Office docs, and RAG files for malware exploits, macros, CVEs, and AI prompt injection. | vendor's own |
| mcp-safeguard cognivators.com | MCP server | MCP server security scanner: detects prompt injection, credential leaks, SSRF, tool poisoning. | vendor's own |
| Overwing overwing.ai | MCP server | Guardrails for LLM output: pass / fail / review verdicts and one recommended action. Hosted or npm. | vendor's own |
| RevClamp MCP Server revclamp.com | MCP server | Zero-take-rate Stripe 2025-03-31.basil, Visa CE 3.0 & EU/XI VIES VAT guardrails MCP server. | vendor's own, widely used |
| SafePrompt safeprompt.dev | MCP server | Detect prompt injection, jailbreaks, and code injection in untrusted text before it reaches an LLM. | vendor's own |
| SkillsSafe Security Scanner skillssafe.com | MCP server | AI skill security scanner. Detects prompt injection, credential theft, ClawHavoc. Free, no signup. | vendor's own |
| Stratum MCP smartmemory.ai | MCP server | Structured execution for coding agents: contracts, postconditions, gates, guardrails. | vendor's own |
How we test this category
A set of prompts with injections, jailbreaks, personal data and clean inputs run through each filter. We count what is caught and what is wrongly blocked, and measure the latency each adds.
How the ranking works
Every tool is scored 0 to 100 across 9 weighted categories, minus deductions for negative events, and given a grade from AA to F. Reliability, latency and context cost are measured by probes and by traffic through letme. Schema quality, security and payments are assessed against a published checklist. Task success comes from a fixed task suite. Payments, including x402, are one category of the nine. Read the methodology.
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.