Grepture

Grepture is an AI gateway that intercepts every LLM request your app makes, rather than observing from the outside. It redacts PII, blocks prompt injections, and enforces rules before data leaves your network. At the same time, it captures full traces, scores responses with built-in evals, and tracks token costs. One integration point handles all of it.

The TypeScript SDK wraps your existing OpenAI, Anthropic, or Gemini client with a single options call. No restructuring your codebase. It also works with Azure OpenAI, Cohere, Mistral, AWS Bedrock, HuggingFace, Groq, and Replicate.

Key capabilities:

• Tracing Full-text search across prompts and responses, waterfall timelines for multi-step agent traces, and request replay with before/after diffs.
• Built-in evals LLM-as-judge scoring runs on live traffic with six templates out of the box. No separate eval pipeline to set up or maintain.
• PII redaction 80+ detection rules identify and redact sensitive fields in-flight. Redaction happens before data reaches any external service.
• Cost tracking Token usage and spend are logged per request, so you can catch runaway costs before users notice.
• Zero-data mode Grepture can process every request without writing content to disk. Only operational metadata (method, status, latency, rule hits) is stored.
• Prompt management Centralized prompt storage accessible across your stack.

For teams with compliance requirements, all infrastructure runs in Frankfurt and Nuremberg. Every subprocessor is EU-based, with no US data transfers. The setup is GDPR-ready by default, which puts it ahead of alternatives like Cloudflare AI Gateway for European deployments.

The gateway is fully open source. Every detection rule and redaction action is readable on GitHub, so there's no black-box processing to audit around. Teams that need complete infrastructure control can self-host. A free tier covers up to 1,000 requests per month.