python/v1.45.0

[gautamsirdeshmukh]

What's Changed (Python SDK)

✨ Features

• feat(memory): support managed Bedrock knowledge bases in retrieve and ACL (shared with TS) by @notowen333 in feat(memory): support managed Bedrock knowledge bases in retrieve and ACL #2909
• feat(offloader): add search/grep support to Python retrieval tool by @lizradway in feat(offloader): add search/grep support to Python retrieval tool #2878
• feat(bidi/openai): support image input via BidiImageInputEvent by @cagataycali in feat(bidi/openai): support image input via BidiImageInputEvent #2327
• feat: implement MCP progress notifications in MCPClient by @joshwand in feat: implement MCP progress notifications in MCPClient #2290
• feat: make BedrockModel._format_request and _convert_non_streaming_to_streaming overridable by @celealramos in feat: make BedrockModel._format_request and _convert_non_streaming_to… #2315
• feat(models)!: support mistralai 2.x in Mistral provider (breaking: requires mistralai>=2.0.0) by @pgrayy in feat(models)!: support mistralai 2.x in Mistral provider #2917

🐛 Fixes

• fix: forward structured output request params for OpenAI Chat Completions by @gautamsirdeshmukh in fix: forward structured output request params for OpenAI Chat Completions #2944
• fix: pass structured output request params by @he-yufeng in fix: pass structured output request params #2396
• fix: support non-streaming OpenAI chat completions by @he-yufeng in fix: support non-streaming OpenAI chat completions #2400
• fix: report OpenAI Responses cached tokens by @fengjikui in fix: report OpenAI Responses cached tokens #2663
• fix(openai-responses): surface cache read tokens in metadata chunk by @gautham18113 in fix(openai-responses): surface cache read tokens in metadata chunk #2555
• fix(openai-responses): prevent tool list mutation across API calls by @xlyoung in fix(openai-responses): prevent tool list mutation across API calls #2551
• fix(anthropic): bypass model_dump on content_block_stop by @chaynabors in fix(anthropic): bypass model_dump on content_block_stop #2953
• fix(gemini): update GeminiModel to handle empty tools in Vertex AI mode by @pshiko in fix(gemini): update GeminiModel to handle empty tools in Vertex AI mode #1040
• fix(ollama): avoid UnboundLocalError on empty model stream by @he-yufeng in fix(ollama): avoid UnboundLocalError on empty model stream #2823
• fix(litellm): generate toolUseId when missing from provider by @chaynabors in fix(litellm): generate toolUseId when missing from provider #2949
• fix: handle empty Bedrock content blocks by @he-yufeng in fix: handle empty Bedrock content blocks #2656
• fix: include reasoning block with empty text and non-empty signature by @gaurav71531 in fix: include reasoning block with empty text and non-empty signature #2150
• fix: preserve non-ASCII tool result text by @cyphercodes in fix: preserve non-ASCII tool result text #2653
• fix: bound tool schema normalization recursion depth by @pgrayy in fix: bound tool schema normalization recursion depth #2853
• fix(tools): unwrap FieldInfo default to avoid pydantic warning by @chaynabors in fix(tools): unwrap FieldInfo default to avoid pydantic warning #2955
• fix(agent): stop idempotency waiters from blocking thread-pool workers by @BV-Venky in fix(agent): stop idempotency waiters from blocking thread-pool workers #2932
• fix(agent): release invocation lock only in throw mode by @chaynabors in fix(agent): release invocation lock only in throw mode #2954
• fix: consume orphaned task exception on stream cancellation by @mehtarac in fix: consume orphaned task exception on stream cancellation #2916

♻️ Refactors

• refactor(memory): type the Bedrock KB store's S3 client via boto3-stubs by @pgrayy in refactor(memory): type the Bedrock KB store's S3 client via boto3-stubs #2847
• refactor(sandbox): inline tool prefixing into getTools implementations (shared with TS) by @gautamsirdeshmukh in refactor(sandbox): inline tool prefixing into getTools implementations #2806

🧪 Tests & Chores

• fix(test): make gemini thinking-model tool-call integ test robust by @yonib05 in fix(test): make gemini thinking-model tool-call integ test robust #2861
• fix(test): retry flaky guardrail output-intervention test by @yonib05 in fix(test): retry flaky guardrail output-intervention test #2869
• fix(test): raise timeout on multimodal swarm/graph integ tests to reduce flakiness by @yonib05 in fix(test): raise timeout on multimodal swarm/graph integ tests to reduce flakiness #2871
• ci(python): add timeouts and apt retries to test-lint workflow by @yonib05 in ci(python): add timeouts and apt retries to test-lint workflow #2874
• ci(python): revert mistralai upper bound to <2.0.0 by @chaynabors in ci(python): revert mistralai upper bound to <2.0.0 #2900
• ci: tighten dependabot versioning strategy by @chaynabors in ci: tighten dependabot versioning strategy #2899
• docs(evals): updated examples to be async by @poshinchen in docs(evals): updated examples to be async #2846

New Contributors

• @fengjikui made their first contribution in fix: report OpenAI Responses cached tokens #2663
• @celealramos made their first contribution in feat: make BedrockModel._format_request and _convert_non_streaming_to… #2315
• @gaurav71531 made their first contribution in fix: include reasoning block with empty text and non-empty signature #2150
• @cyphercodes made their first contribution in fix: preserve non-ASCII tool result text #2653
• @gautham18113 made their first contribution in fix(openai-responses): surface cache read tokens in metadata chunk #2555
• @xlyoung made their first contribution in fix(openai-responses): prevent tool list mutation across API calls #2551
• @joshwand made their first contribution in feat: implement MCP progress notifications in MCPClient #2290

Full Changelog: python/v1.44.0...python/v1.45.0

---

This discussion was created from the release python/v1.45.0.