strands-agents

This project publishes no release notes. Between v1.40.0 and v1.41.0 there were 4 commits, 4 of them substantive:

• feat(plugins): add MultiAgentPlugin for Swarm and Graph orchestrators (#2280)
• feat: bump starlette dependency to 1.x (#2297)
• feat(bedrock): add TTL support to auto-injected tool and system/user cache points (#2232)
• fix(tests): add use_native_token_count=True when expected (#2311)

This project publishes no release notes. Between v1.39.0 and v1.40.0 there were 6 commits, 6 of them substantive:

• feat: add proactive context compression to conversation managers (#2239)
• feat: cache AccessDenied error for count tokens (#2279)
• fix(ollama): update return type of latencyMs metric for ollama model provider (#2236)
• feat: add official Discord link (#2285)
• fix: set use_native_token_count default to false (#2284)
• fix: swarm bug "Failed to detach context" with opentelemetry (#2281)

This project publishes no release notes. Between v1.38.0 and v1.39.0 there were 9 commits, 9 of them substantive:

• feat: enable openai provider use aws profile (#2230)
• fix: include root cause in MCPClientInitializationError message (#2238)
• feat: add context window limit lookup table (#2249)
• fix: fix count tokens for bedrock models (#2254)
• fix: cache unsupported models for bedrocks token counting (#2250)
• feat: add useNativeTokenCount flag to skip token counting API calls (#2255)
• fix: correct MCPClient.exit and stop() type annotations (#2248)
• feat(a2a): implement full A2A task lifecycle state support (#2245)
• fix: integration test updates (#2262)

This project publishes no release notes. Between v1.37.0 and v1.38.0 there were 22 commits, 19 of them substantive:

• feat(mcp): preserve CallToolResult.isError flag in MCPToolResult (#2118)
• feat: add count_token method to model with naive estimation using tiktoken (#2031)
• chore(log): added warning for default model awareness and is subject to change (#2164)
• fix(litellm): forward ttl field from CachePoint in _format_system_messages (#2153)
• fix(skills): preserve cache points in system prompt during skills inj… (#2134)
• fix(ollama): generate unique toolUseId instead of reusing tool name (#2053)
• feat(cache): add TTL support to CachePoint for prompt caching (#1660)
• fix: use non-interactive flag for Nova Sonic history and system promp… (#2188)
• feat: large tool result offload (#2162)
• feat: override count_tokens with native token counting for supported providers (#2189)
• fix(bedrock): upgrade default model to Claude Sonnet 4.5 (#2193)
• feat: add ProviderTokenCountError for native token counting failures (#2211)
• fix(conversation-manager): handle window_size=0 and reject negative values (#2208)
• fix: change token counting fallback log from warning to debug (#2220)
• fix: do not synthesize exception for cancelled tools (#2106)
• feat: estimate input tokens before model calls (#2221)
• feat(offloader): return explicit paths in preview and auto-enable retrieval (#2222)
• fix: update tests to use non-EOL'd model (#2226)
• feat(bedrock): add strict_tools config with auto-inject of additional… (#2213)

This project publishes no release notes. Between v1.36.0 and v1.37.0 there were 5 commits, 5 of them substantive:

• fix: add fallback trim point for tool-heavy conversations in SlidingWindowConversationManager (#2174)
• feat: introduce checkpoint in experimental (#2181)
• feat: add context_window_limit to model configs (#2176)
• fix(mcp): skip MCPClient cleanup during interpreter finalization (#2144)
• fix(tests): update retired claude-3-haiku model in integration tests (#2186)

This project publishes no release notes. Between v1.35.0 and v1.36.0 there were 14 commits, 14 of them substantive:

• feat(hooks): accept callable hook callbacks in Agent constructor (#1992)
• fix: handle missing optional fields in non-streaming citation conversion (#2098)
• fix(telemetry): add common gen_ai attributes to event loop cycle spans (#1973)
• fix(telemetry): use per-invocation usage in agent span attributes (#2017)
• feat(a2a): add client_config param and deprecate a2a_client_factory (#2103)
• fix: clear leaked running loop in MCP client background thread (#2111)
• feat(openai): plumb through cache tokens in metadata events (#2116)
• feat(agent): add take_snapshot() and load_snapshot() methods (#1948)
• feat(skills): support loading skills from URLs (#2091)
• feat: add metadata field to messages for stateful context tracking (#2125)
• feat(bidi): support request_state stop_event_loop flag (#1954)
• fix: preserve Gemini thought_signature in LiteLLM multi-turn tool calls (#2129)
• fix(bedrock): normalize empty toolResult content arrays in _format_bedrock_messages (#2123)
• fix(telemetry): remove force_flush in tracer (#2142)

This project publishes no release notes. Between v1.34.1 and v1.35.0 there were 11 commits, 11 of them substantive:

• feat: add service_tier support to BedrockModel (#1799)
• fix: forward _meta to MCP tool calls and fix model_dump alias seriali… (#1918)
• fix(anthropic): avoid Pydantic warnings for message_stop events (#2044)
• fix: propagate tool exceptions to spans so StatusCode.ERROR is set correctly (#2046)
• fix(docs): update 19 broken documentation links in README (#1906)
• fix: enforce that the first message is a user message in the sliding window conversation manager (#2087)
• fix: forward meta to MCP task-augmented tool calls (#2081)
• fix: handle premature stream termination for Anthropic (#1868) (#2047)
• ci: add weekly markdown link check workflow (#2088)
• fix(test): update session integ test for sliding window conversation manager (#2092)
• fix(test): fix anthropic stream test mock missing get_final_message (#2094)

This project publishes no release notes. Between v1.34.0 and v1.34.1 there were 4 commits, 4 of them substantive:

• fix: fix type imcompatible (#2018)
• fix: isolate langfuse env vars (#2022)
• fix: restore explicit span.end() to fix span end_time regression (#2032)
• feat(context): surface context window size from LLM response metadata

This project publishes no release notes. Between v1.33.0 and v1.34.0 there were 9 commits, 9 of them substantive:

• chore: remove Cohere from required integ test providers (#1967)
• feat: add AgentAsTool (#1932)
• feat: auto-wrap Agent instances passed in tools list (#1997)
• feat(telemetry): emit system prompt on chat spans per GenAI semconv (#1818)
• feat(mcp): add support for MCP elicitation -32042 error handling (#1745)
• fix: ollama input/output token count (#2008)
• feat: add stateful model support for server-side conversation management (#2004)
• feat: add built-in tool support for OpenAI Responses API (#2011)
• fix: handle reasoning content in OpenAIResponsesModel request formatting (#2013)