← Guardrails

DeepSeek Guardrails

Guardrails for DeepSeek agents. Route through DataHippo and enforce policies on every call.

Supported Models

DeepSeek V3, DeepSeek R1

Features

  • Block prompt injection on DeepSeek models
  • Screen outputs for data leakage
  • Custom content policies
  • Audit logging

Setup

Point your DeepSeek client at DataHippo:

https://flow.datahippohq.com/deepseek/v1/chat/completions
Screenshot

Five layers of protection

Every DeepSeek request passes through all five guardrail layers before it reaches the model. Same protection. Zero code changes.

🔒

PII Detection

Emails, phone numbers, and API keys are stripped from prompts and redacted in responses. Automatic. No config needed.

🛡

Prompt Injection Detection

Known injection patterns are blocked at the gateway. Suspicious inputs are flagged for review. Your model never sees them.

🚫

Topic Blocking

Define off-limit subjects per project and per agent. Competitor names, confidential data, restricted topics. Enforced automatically.

Whitelisted Tools

Only approved functions can be called. No unauthorized API calls. No surprises. You define the boundary.

💰

Cost Caps

Set spend limits per session, per agent, per project. Requests are rejected at the cap. No runaway costs.