v3.17.0

• LangChain instrumentation via LangSmith by @alexmojaki in #1084
• MCP OTel context propagation by @samuelcolvin in #1103
• Add capfire.get_collected_metrics() by @alexmojaki in #1116

v3.16.2

• Fixes for OpenAI Responses API and Agents SDK by @alexmojaki in #1092, #1093, #1094, and #1095
• Fix verbose console formatting for enum, dates, and decimals by @sbhrwlr in #1096
• Allow setting logfire.msg in structlog integration by @alexmojaki in #1113
• Add ASGI instrumentation package to django extra by @alexmojaki in #1097

v3.16.1

• Infer base URL from read token in query client by @Viicos in #1088
• Add include_binary_content (#1090) and **kwargs (#1078) to instrument_pydantic_ai by @alexmojaki

v3.16.0

• Make OpenAI spans show token usage in logfire UI by @alexmojaki in #1076
• Fixes for verbose console logging by @alexmojaki in #1071 and #1072
• Export first batch of spans more quickly by @alexmojaki in #1066
• Tighten scrubbing patterns to reduce accidental matches by @alexmojaki in #1074
• Add do_not_scrub and binary_content as safe keys for scrubber by @alexmojaki in #1075

v3.15.1

• Support OpenTelemetry SDK 1.33.0 by @alexmojaki in #1067

v3.15.0

• Remove attributes from http.server.active_requests metric to prevent emitting too many by @alexmojaki in #1060
  • This is technically a breaking change as it means less data is sent to Logfire, but most users don't use it and some will save a significant amount of money.

v3.14.1

• Handle changes in openai and anthropic by @alexmojaki in #1030
• Fix exporting of very large spans and payloads by @alexmojaki in #1027
• Prevent infinite loop in get_user_frame_and_stacklevel by @alexmojaki in #1031

v3.14.0

• Experimental functions for recording feedback annotations

v3.13.1

• Upgrade to OpenTelemetry SDK 1.32.0 by @alexmojaki in #991

v3.13.0

• Emit logs sent from MCP server to client by @alexmojaki in #974
• Return None from logfire_api.LogfireSpan.context when logfire could not be imported by @DouweM in #983