What's Changed
- fix(agenticclaude): align direct anthropic auth precedence with sdk behavior by @hi-pender in #884
- fix(semantic): correct split by @meguminnnnnnnnn in #887
- fix(callbacks): fix cozeloop token usage by @shentongmartin in #897
- feat(claude): add structured output support by @marwan38 in #709
- docs(agentic): enrich cache section with mermaid flowchart and examples by @mrh997 in #898
- chore(openai): enrich reasoning effort enums by @N3kox in #899
- feat(claude): add adaptive thinking support by @jesse-engineer in #902
- feat(libs/acl/langfuse): expose max event size option by @jesse-engineer in #910
- feat(callbacks/langfuse): expose max event size config by @jesse-engineer in #912
- feat(model): support mid-conversation system messages by @mrh997 in #914
- feat(claude): support Vertex service account JSON credentials by @shaikustin in #888
- feat(claude): add RequestTimeout config and per-request option by @mrh997 in #920
- fix(deepseek): map completion_tokens_details.reasoning_tokens in usage by @jesse-engineer in #921
- fix(claude): map output_tokens_details.thinking_tokens in usage by @jesse-engineer in #927
- feat(officialmcp): reconnecting session via ClientSession anti-corruption interface by @hi-pender in #889
- feat(claude): expose cache_creation_input_tokens in message.Extra and add Config.AutoCacheControl by @sin-bufan in #929
- fix(ark): support fast service tier in responses API by @Abtiotm in #938
- feat(officialmcp): support custom transport factories by @fanlv in #946
- fix(tool_mcp): fix InvokableRun crash on empty arguments #923 by @kayoch1n in #924
- feat(callbacks/langfuse): add v2 OTLP callback by @jesse-engineer in #954
- fix(claude): agenticclaude panic when ToJSONSchema returns nil, nil by @N3kox in #956
- fix(agenticclaude): preserve complete tool JSON schemas by @hi-pender in #971
- chore: upgrade Gemini GenAI SDK to v1.70.0 by @mrh997 in #979
- feat(agenticclaude): expose cache_creation_input_tokens in message.Extra by @sin-bufan in #978
- feat(langfuse): expose OTLP retry config by @jesse-engineer in #981
- fix(ark): support EnhancedTool text results in Responses API by @N3kox in #986
- fix(agenticark): preserve raw reasoning content from Ark Responses API by @N3kox in #985
- fix(langfuse): preserve agent outcomes through retries and callback panics by @jesse-engineer in #989
- feat(gemini): add WithLabels option for request-level metadata by @jredl-va in #769
- fix(gemini): preserve context errors during streaming by @jesse-engineer in #987
- feat(callbacks/tls): add opt-in TLS observability callback by @runninggo in #936
- feat: add agentic response modifiers to OpenAI ACL by @mrh997 in #1010
- fix: isolate OpenAI request extra fields by @mrh997 in #1012
- fix: finalize agentic output before callbacks by @mrh997 in #1011
New Contributors
- @marwan38 made their first contribution in #709
- @shaikustin made their first contribution in #888
- @sin-bufan made their first contribution in #929
- @Abtiotm made their first contribution in #938
- @jredl-va made their first contribution in #769
- @runninggo made their first contribution in #936
Full Changelog: components/model/agenticark/v0.2.4...components/model/agenticark/v0.2.5