How do I secure Replicate applications?

Integrate Wardstone Guard with your Replicate application to detect prompt injections, jailbreaks, harmful content, and data leakage. Wardstone validates inputs before they reach Replicate and screens outputs before returning them to users.

Replicate + Wardstone

Secure Cloud Model Hosting

Protect your Replicate model deployments with Wardstone Guard. Secure open-source models running on Replicate's infrastructure from prompt attacks.

15 Supported Models· all protected with sub-30ms latency

Llama 4 ScoutLlama 4 MaverickLlama 3.3 70BLlama 3.2 VisionDeepSeek-V3DeepSeek-R1+9 more

Try the Playground Visit Replicate

Why Secure Replicate?

No Provider Safety Layer

High Risk

Replicate runs models without modifying them, including any safety gaps.

Image Generation Abuse

High Risk

Stable Diffusion and other image models can generate harmful visual content.

Model Version Inconsistency

Medium Risk

Different model versions may have different vulnerability profiles.

Security Considerations

1
Replicate hosts models as-is without additional safety layers
2
Prediction pricing means security should be lightweight
3
Image models can generate harmful visual content
4
Wardstone protects both text and describes image generation prompts

How to Integrate

Add Wardstone to your Replicate calls
Install Wardstone SDK alongside the Replicate client.
Validate prediction inputs
Screen prompts before calling replicate.run() or predictions.create().
Screen text outputs
Validate text model outputs for harmful content.
Implement prompt policies
Block harmful image generation prompts before they reach Stable Diffusion.

Pricing Note

Replicate charges per prediction. Wardstone's lightweight inference adds minimal latency and cost.

Secure Replicate with Wardstone

# Step 1: Check user input with Wardstone
curl -X POST "https://api.wardstone.ai/v1/detect" \
  -H "Authorization: Bearer YOUR_WARDSTONE_KEY" \
  -H "Content-Type: application/json" \
  -d '{"text": "User message here"}'
 
# Response: { "prompt_attack": { "detected": false, ... } }
 
# Step 2: If safe, send to Replicate
curl -X POST "https://api.replicate.com/v1/predictions" \
  -H "Authorization: Bearer YOUR_REPLICATE_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "version": "meta/llama-4-scout-109b-instruct",
    "input": {"prompt": "User message here", "max_tokens": 500}
  }'
 
# Step 3: Check Replicate response with Wardstone before returning to user

Common Use Cases

Rapid prototyping

Pay-per-inference deployments

Open-source model access

Image generation applications

Model comparison and testing

All Supported Replicate Models

Wardstone Guard protects all Replicate models with the same comprehensive security coverage. Whether you're using the latest releases or legacy models still in production, every API call is protected.

Llama 4 Scout

Llama 4 Maverick

Llama 3.3 70B

Llama 3.2 Vision

DeepSeek-V3

DeepSeek-R1

Mistral Large

FLUX.1 Pro

FLUX.1 Schnell

FLUX.1 Dev

Stable Diffusion 3.5

SDXL