PROXY ENGINE Injection Filter Content Policy Rate Limiter Audit Logger Cost Controls API Compat Request LLM API user prompt ignore previous instructions BOUNDARY LAYER 3 BOUNDARY LAYER 2 BOUNDARY LAYER 1 Safe output

Prompt hardening for LLM-powered applications

A drop-in proxy that hardens your LLM prompts against injection attacks — 99.5%+ effective, no code changes, no extra API calls. Stop writing defensive boilerplate and get back to building product.

Prompt Hardening Proxy

Platemail Proxy

A transparent proxy that hardens every prompt before it reaches the model. Nonce-based role markers, attention dilution, tag defanging, and secret scanning — compiled from YAML policy, enforced at runtime.

  • Drop-in for OpenAI, Anthropic, and any OpenAI-compatible API
  • Unforgeable nonce markers and attention dilution against prompt injection
  • Context window advertisement — clients self-limit, proxy always has room
  • Aho-Corasick secret scanning with encoding-aware evasion detection
  • Extensible hook system — PII detection, content filtering, LLM judge
  • OTEL trace annotations for existing observability stacks
View on GitHub How It Works → For Teams →

Get in touch

Questions, partnerships, or security disclosures.

hello@platemail.ai