For side projects and proofs of concept.
- 10,000 protected requests / month
- All 6 defense layers
- 1 app · 1 seat
- 7-day log retention
- Community support
- Custom policies
- SSO / SAML
Start free, upgrade when you grow. Every plan includes all six defense layers — because security shouldn't be an add-on.
For side projects and proofs of concept.
For teams shipping AI features to production.
For regulated industries and serious scale.
One prompt-and-response cycle that flows through the PromptShield gateway. Cached or blocked-before-model requests still count, but at 1/10th weight.
Median overhead is under 12 ms on Growth and Enterprise plans thanks to regional edge scanning. You can also run async “shadow mode” with zero inline latency.
OpenAI, Anthropic, Google Gemini, Azure OpenAI, AWS Bedrock, and any self-hosted model behind an HTTP endpoint. The SDK is model-agnostic.
Never by default. Blocked-attack metadata is retained per your plan's window; raw prompt bodies are only stored if you explicitly enable safe-copy logging (with PII redaction applied first).
Yes — shadow mode mirrors your traffic and reports what PromptShield would have blocked, with zero risk to live users. Most teams start there.
Tell us your request volume and compliance needs — we'll recommend the smallest plan that keeps you safe.