Content filtering, PII detection, prompt injection protection, and safety controls.
Built-in guardrail count: Corveil ships with 6 built-in guardrail plugins (content moderation, PII filter, jailbreak detector, anonymizer, cost alerter, keyword blocklist) plus unlimited custom guardrails configurable via API. Portkey offers the largest guardrail library with 60+ options including 40+ pre-built guardrails and partner integrations. LiteLLM includes basic regex and keyword guardrails built-in; advanced guardrails require third-party paid services like Presidio or PANW Prisma. Cloudflare uses Meta Llama Guard with fixed content safety categories that are not customizable or pluggable. Helicone offers only OpenAI moderation API passthrough with no additional guardrail capabilities. Bifrost has no built-in guardrails and delegates all content filtering to third-party services like AWS Bedrock Guardrails and Azure Content Safety. Kong offers PII sanitizer and semantic prompt guard plugins, but both require Enterprise licensing.
PII detection and handling: Corveil detects SSN, credit card, email, phone, and IP patterns with three handling modes: block, redact, or anonymize-and-restore (strips PII before the LLM call, restores real values in the response). Portkey can detect and redact PII but does not offer round-trip anonymization with restoration. LiteLLM PII handling requires integration with third-party services like Presidio, Lasso, or PANW Prisma. Cloudflare detects PII via DLP profiles and can block or alert, but does not offer redaction or anonymization. Helicone does not offer PII detection or handling. Bifrost has no built-in PII detection and relies entirely on third-party services. Kong’s AI Sanitizer plugin supports 20 PII categories in 9 languages but requires Enterprise licensing and does not offer round-trip anonymization.
Jailbreak / prompt injection detection: Corveil includes a built-in jailbreak detector with 8+ default patterns plus custom regex rules, configurable via API at runtime. Portkey includes jailbreak and prompt injection detection as part of its guardrail library. LiteLLM supports basic regex and keyword-based prompt injection detection. Cloudflare does not offer jailbreak or prompt injection detection as a distinct feature. Helicone offers LLM-based injection detection but no pattern-based screening. Bifrost has no built-in jailbreak detection. Kong offers regex and semantic prompt guard plugins on Enterprise tier only.
Custom guardrails via API: Corveil supports creating, testing, enabling, and disabling custom guardrails via REST API at runtime without redeployment. Portkey supports bring-your-own-guardrails with API configuration. LiteLLM guardrails are configured in code or YAML, not via runtime API. Cloudflare guardrails are limited to fixed Llama Guard categories with no custom configuration. Helicone does not support custom guardrail configuration. Bifrost has no guardrail system to configure. Kong guardrail configuration requires updating plugin settings, not a dedicated guardrails API.