Wardstone

What types of threats does Wardstone detect?

Wardstone detects four categories of threats: prompt attacks (jailbreaks, injections), harmful content (violence, hate speech, sexual content), data leakage (PII like SSNs, credit cards, emails), and unknown links (suspicious URLs not on our allowlist). All categories are evaluated in a single API call.

How fast is the API response time?

Our P95 latency is Sub-30ms. Wardstone uses optimized ONNX inference that runs on CPU, so you don't need specialized hardware. This makes it fast enough to run synchronously in your request pipeline without noticeable delay to end users.

Does Wardstone work with any LLM provider?

Yes, Wardstone is completely model-agnostic. It works with OpenAI, Anthropic, Google, Meta, Mistral, and any other LLM provider. Since we analyze text content rather than model internals, you can use us with any AI system.

What's included in the free tier?

The free tier includes 10,000 API calls per month, access to all four detection categories, playground access for testing, and community support. No credit card required to start.

How does pricing work for overages?

On the Business plan, overages are billed at $0.50 per 1,000 additional API calls. There are no surprise charges: you can set usage alerts and hard limits in your dashboard. Enterprise customers get custom volume pricing.

Can I use Wardstone for both input and output scanning?

Absolutely. Wardstone provides bidirectional protection. Scan user inputs before they reach your LLM to block attacks and injections, then scan model outputs before delivery to catch harmful content or data leakage.

Is there an on-premise deployment option?

Yes, on-premise deployment is available for Enterprise customers. This includes the full Guard model, custom SLAs, and dedicated support for integration. Contact our sales team to discuss your requirements.

How do I integrate Wardstone into my application?

Integration takes minutes. We offer a REST API and SDKs for popular languages. Call POST /api/detect with your text and read flagged, risk_bands, and primary_category in the response. Paid plans can optionally include raw model scores.