How do I secure Together AI applications?

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

Together AI + Wardstone

Secure Open Models at Scale

Add security to Together AI's lightning-fast inference. Protect Llama, Mistral, DeepSeek, and other open models from prompt attacks and data leakage at any scale.

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

Llama 4 ScoutLlama 4 MaverickLlama 3.3 70B TurboLlama 3.2 90B VisionLlama 3.1 405B TurboDeepSeek-V3+9 more

Try the Playground Visit Together AI

Why Secure Together AI?

Minimal Native Safety

High Risk

Together focuses on inference speed, not safety filtering.

Fine-Tuned Model Risks

High Risk

Custom models hosted on Together may have compromised safety training.

Scale Attack Surface

Medium Risk

Fast serverless inference can amplify successful attacks rapidly.

Security Considerations

1
Together provides fast inference but minimal safety filtering
2
Custom fine-tuned models may have degraded safety
3
Serverless inference means rapid scaling of potential attacks
4
Wardstone handles security so you can focus on inference optimization

How to Integrate

Install Wardstone SDK
Add Wardstone to your Together AI integration.
Pre-screen all requests
Validate inputs before calling Together's chat completion API.
Post-process outputs
Screen model responses for harmful content and data leakage.
Apply consistent policies
Use the same security policy across different Together models.

Pricing Note

Together offers competitive open-source model pricing. Wardstone adds security without negating cost benefits.

Secure Together AI 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 Together AI
curl -X POST "https://api.together.xyz/v1/chat/completions" \
  -H "Authorization: Bearer YOUR_TOGETHER_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "meta-llama/Llama-4-Scout-17B-16E-Instruct",
    "messages": [{"role": "user", "content": "User message here"}]
  }'
 
# Step 3: Check Together AI response with Wardstone before returning to user

Common Use Cases

Production open-source deployments

Cost-optimized inference

Fine-tuned model hosting

Multi-model orchestration

Enterprise open-source adoption

All Supported Together AI Models

Wardstone Guard protects all Together AI 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 Turbo

Llama 3.2 90B Vision

Llama 3.1 405B Turbo

DeepSeek-V3

DeepSeek-R1

Qwen2.5 72B

Qwen2.5 Coder 32B

Mistral Large

Mixtral 8x22B

FLUX.1 Pro