regrada

A continuous integration gate for large language model behavior that records live LLM traffic through an HTTP proxy, converts traces into version-controlled test cases, and enforces policies in CI. It solves the problem of silent behavioral regressions in AI applications that unit tests miss. It sells to software developers, engineering teams, and DevOps or SRE teams in B2B organizations. It is delivered as a SaaS with an API and CLI, requiring no code changes, and offers a free tier and enterprise pricing.

Key features

• Zero-code traffic capture via HTTP proxy
• Convert traces to YAML test cases
• Policy engine for assertions and thresholds
• Automatic PII and secrets redaction
• Native GitHub Actions integration
• Web dashboard for trace history
• Support for OpenAI, Anthropic, Azure, Bedrock
• Built-in mock provider for local tests