py-sdk-v0.4.0

Changelog

• Fix global scoping (generated types update) (#1213)
• py sdk and integrations now require a min. of Python 3.10 (#1194)

js-sdk-v1.1.1

Changelog

• Fix improper handling of async generator as execute input (#1203)

py-sdk-v0.3.15

Changelog

• wrap_openai now sends smaller logs with braintrust attachments (#1180)
• New Pydantic AI integration setup_pydantic_ai (#1192)
• Small improvement to function invocation context (#1167)

js-sdk-v1.1.0

Changelog

• wrapOpenAI now produces smaller logs with braintrust attachments (#1180)
• Support AI SDK v6 new metrics objects (#1182)
• Improve function invocation context (#1167)
• Add ability to create prompts with nunjucks template format (#1193, #1075)
• Use Vercel's AI Gateway's cost, if available, in AI SDK (#1166)

py-sdk-v0.3.14

Changelog

• Fix UTF-8 encoding for non-ASCII log data with orjson + urllib3 1.x (#1183)
• Fix Score.as_dict() to include name field (#1103)
• Make sure ancestor commits never exceed the specified limit (#952)

js-sdk-v1.0.3

Changelog

• Evals called directly (outside of using the CLI) will now return plain text (no color, etc.). This avoids additional dependencies that break specific builds/environments. (#1165)
• browser entrypoint has 1:1 parity with main (nodejs). This adds Eval and a few SDK integrations that were not previously included due to non-isometric reasons. (#1165)
• The cli entrypoint is now cjs only. Imports attempts from the main SDK code to the CLI code will be denied to prevent coupling.
• We now prevent public interfaces/types from regressing within major versions. (#1172)
• Make sure ancestor commits never exceed the specified limit (#952)
• New Jest smoke test to prevent regressing code tested that has tracing instrumented. (#1160)
• CVE dependency update. (#1164)
• CVE dependency update. (#1162)

py-sdk-v0.3.13

Changelog

• Fetching datasets now support an optional batch size (#1150)
• Improve serialization performance, if orjson is available (#1100)

js-sdk-v1.0.2

Changelog

• Add automatic tracing to OpenAI's streaming structured output (#1158)
• Fetching datasets now have an optional batch size (#1150)
• Add automatic tracing for AI SDK's doGenerate and doStream (#1144)
• Add Agent (and AgentTool) automatic tracing for AI SDK (#1065)
• Fix withResponse() on wrapped OpenAI responses API (#1145)

py-sdk-v0.3.12

Changelog

• Allow and use project id header in remote evals for project-level ai secrets (#1143)
• New Braintrust SDK 1.x (#1112)
• Replace mock unit test with real OpenRouter integration tests (#1136)

js-sdk-v1.0.1

Changelog

• Allow and use project id header in remote evals for project-level ai secrets (#1143)
• Allow links to be clicked in supported terminals (#1139)
• Update package-locks to resolve dependabot issues (#1141)