Skip to main content
v1.4.11
Breaking changes in v1.4.0. See the v1.4.0 Migration Guide for full before/after examples, automatic migration details, and a step-by-step checklist before upgrading.

Changelog

A profiling and provider-fidelity release on transports/v1.5.15. The headline is an optional pprof profiling server, now wired end-to-end into the enterprise binary so you can capture runtime CPU, heap, block, and mutex profiles in production behind a single env flag. This release also surfaces Anthropic prompt-cache diagnostics so you can see exactly where a cache prefix diverged, and tightens native GenAI request passthrough and tool-call metadata preservation for Gemini/Vertex.

✨ Features

  • pprof Profiling Server (OSS + enterprise) - Optional runtime profiling server gated by BIFROST_PPROF_PORT, with env-tunable block and mutex sampling rates and graceful shutdown alongside the main server. The enterprise binary now wires the profiling server into its startup path so the same diagnostics are available in enterprise deployments.
  • Anthropic Cache Diagnostics (OSS) - Surfaces Anthropic’s prompt-cache diagnostics beta (cache-diagnosis-2026-04-07): responses now expose the first cache-prefix divergence point, so you can see exactly why a prompt cache missed.

🐞 Fixed

  • GenAI Raw Request Passthrough (OSS) - Native Vertex/Gemini batch and request bodies now follow the x-model-provider header and pass through verbatim only when Gemini or Vertex is explicitly selected, preventing a raw body from reaching a mismatched provider.
  • Tool Call Metadata Preservation (OSS) - extra_content on assistant tool calls (e.g. Gemini thought_signature) is now preserved across both streaming and non-streaming responses.

📀 Base OSS version

transports/v1.5.15

🔌 If you are compiling plugin against this release - use following deps