Enkrypt AI Guardrails

Overview

The product can approve, change, or block behavior at prompt, retrieval, tool, and output boundaries. That matters in agent workflows, where risk can arise in a tool action or retrieved data as well as in a final answer. Coverage spans prompt injection, jailbreaks, unsafe tool actions, privilege or tenant boundary violations, sensitive data exfiltration, toxic content, and compliance risks.

Policies also cover image and audio inputs, including injection defenses. This makes Enkrypt relevant to multimodal deployments, not just text-based applications. It is a runtime control layer, though, so teams need to fit its enforcement into the paths their own models and agents use.

Key features

Controls and integration

Filtering, rewriting, blocking, escalation, and tool-call approval or denial give teams several ways to respond to policy violations. API-first use, agent hooks or middleware, and an MCP Gateway offer options for connecting the controls to an existing stack. Identity claims from Okta, Azure AD, or custom JWT claims can inform policy decisions where access context matters. Python is the listed SDK language, which may mean extra integration work for teams building in other languages.

Audit and performance

Enforcement logs include a policy ID, version, and reason code, and can be exported to a SIEM. That gives security teams useful evidence for internal control reviews. The product page claims latency under 15 milliseconds and stable behavior under load; those claims are relevant to runtime use, but do not replace evaluation against a deployment's own requirements.

The pricing page identifies SOC 2 Type II, ISO 27001, GDPR Ready, HIPAA Ready, and NIST AI RMF Aligned. These credentials and alignments may help enterprise review, while teams should match them to their own compliance needs.

Pricing

Explore is free forever at 0.00 USD per free. It starts with 500 credits, then includes 50 credits per month, 7-day data retention, and community support. That works for spot checks and evaluation, but the 50-credit monthly allowance is modest: a comprehensive red team assessment requires 5,000 credits.

Launch costs 149.00 USD per month, billed monthly, with 5,000 credits to start, 250 credits per month, 30-day retention, and email support. Its starting allocation matches the stated threshold for a comprehensive red team assessment, but the recurring monthly quota is much smaller. Scale costs 1499.00 USD per month, billed monthly, and provides 10,000 credits to start, 1,000 per month, 30-day retention, and dedicated Slack or Teams support. It suits teams with greater ongoing usage and a need for closer support, though its monthly allocation remains finite.

Enterprise has custom pricing, custom allocation, custom retention, custom SLAs, and unlimited usage for VPC deployment. It is the fit for organizations needing VPC or on-premises deployment or terms beyond the fixed plans. The pricing note gives a paid-from figure of 134/mo (annual); the stated Launch and Scale prices are billed monthly. Compare the applicable term and credit needs before choosing.

Platforms

Enkrypt supports API, self-hosted, and web platforms. Enterprise includes VPC or on-premises deployment, with unlimited usage for VPC.

Who it's for

Enkrypt is aimed at enterprises securing AI deployments, AI safety researchers, and people building safer AI. It is particularly suitable for teams with agent, tool, retrieval, or multimodal risks that need policy decisions recorded for review. It is less compelling for a small evaluation that needs sustained high-volume usage: Explore's monthly allowance is limited, while the paid plans have defined credit quotas.

Pros and cons

• Pros: Controls span prompts, retrieval, tools, and outputs, addressing risks that a response-only filter would not cover.
• Pros: Multimodal policy coverage and several integration paths suit varied AI workflows.
• Pros: Logs with policy identifiers and reason codes can be exported to SIEM for internal evidence.
• Cons: Explore's 50 monthly credits and 7-day retention constrain ongoing evaluation and review.
• Cons: Launch and Scale have finite monthly credits, while custom allocation and retention require Enterprise.
• Cons: Python is the listed SDK language, potentially adding integration effort for other-language teams.

Alternatives

AI Guardrail Software is a useful starting point for comparing products in the category. OpenSecureAI Prompt Firewall may suit users seeking a free plan with in-browser tools and npm packages, 2,000 API requests per month, and a BYOK LLM Gateway. PromptGuard offers a free plan with 20,000 scans a month, one API key, one project, and 24-hour log retention; choose it when scan volume and a broad range of platform support matter more than Enkrypt's stated workflow controls.

Fiddler Guardrails is another free-plan option, with real-time detection for harmful exposure, hallucinations, toxicity, PII/PHI, prompt injection, and jailbreak attempts. Openlayer Guardrails may fit teams that want a free starting plan with one member, five projects, and 20,000 inferences per month. Amazon Bedrock Guardrails is a paid option without a free plan, for readers considering its listed content-filter and denied-topic pricing.

VotalAI is another paid option with a free plan and trial. PII Firewall may suit Linux or self-hosted users looking for a free Apache 2.0 Python package with a regex-only install. HiddenLayer AI Runtime Security is a paid option for readers considering runtime security across its supported platforms.

Verdict

Choose Enkrypt AI Guardrails if your organization needs runtime policy enforcement across agents, tools, retrieval, and multimodal inputs, with auditable decisions that can feed a SIEM. Its breadth is the reason to choose it. Look elsewhere if your priority is a generous recurring free quota or you need to avoid fixed monthly credit limits; those constraints make the lower plans better for bounded evaluation than sustained high-volume use.