Python Agent version notes-Application Real-Time Monitoring Service(ARMS)-阿里云帮助中心

This topic describes the version history of the ARMS Application Monitoring Python Agent.

View agent versions

Log on to the ARMS console.
In the navigation pane on the left, choose Application Monitoring > > Agent Management, and then click the Agent Release Notes tab.

Agent versions

Version	Release date	Release notes
2.9.0	June 12, 2026	New features Support for Qwen-Agent, QwenPaw, and Hermes frameworks. For more information, see Python components and frameworks supported by ARMS Application Monitoring. LangChain plugin now supports Rerank call observability. Enhanced observability for the SGLang inference engine. For SGLang v0.5.9 or later, fine-grained performance tracing is supported, including timing details for tokenizer, scheduler, Prefill, and Decode components and phases. For more information, see Observing vLLM/SGLang inference engines. Fixes Fixed span state crosstalk in AgentScope ReAct multi-turn calls, nested agents, and concurrent scenarios. Fixed inaccurate span end timing in Google ADK streaming calls and concurrent calls within the same session. Fixed GenAI semantic data collection for AI Agent frameworks such as CrewAI, Google ADK, and LiteLLM. Fixed unintended inclusion of ARMS health check requests in business trace data.
2.8.3	June 4, 2026	New features Support for custom metrics using the OpenTelemetry Python SDK. Fixes Fixed erroneous stack traces printed to user stdout during system metric collection in certain scenarios. Fixed metric dimension divergence in multi-process scenarios. Fixed missing output content collection in some DashScope plugin scenarios.
2.7.0	May 18, 2026	New features Support for AI Agent–level observability in Cloud Monitor 2.0’s AI Agent monitoring. For more information, see Cloud Monitor 2.0. Integration with AI Profiling in ECS environments. Support for PD-separation scenarios in observing vLLM/SGLang inference engines. Fixes Fixed missing error metric collection for OpenAI plugin when LLM calls fail. Fixed slow agent startup caused by network issues in certain scenarios.
2.6.4	April 21, 2026	New features vLLM plugin now supports token-level analysis. For more information, see Observing vLLM/SGLang inference engines. Added support for Python 3.14 and ended future support for Python 3.8. For more information, see Python components and frameworks supported by ARMS Application Monitoring. Fixes Fixed missing first-packet latency metric reporting for AgentScope framework. Improved compatibility for collecting session_id and user_id association in LangGraph. Fixed missing gen_ai.provider.name attribute in OpenAI plugin Embedding spans. Fixed server request crashes in GEVENT scenarios.
2.5.4	April 2, 2026	New features Added support for Dify ReACT plugin. For more information, see Python components and frameworks supported by ARMS Application Monitoring. LangChain, AgentScope, and Dify ReACT now support observability for each ReACT iteration. For more information, see LLM Trace field definitions. vLLM plugin now supports engine concurrency analysis and token-level fine-grained observability. For more information, see Observing vLLM/SGLang inference engines. vLLM engine now supports observability for Embedding and Rerank operations. For more information, see Observing vLLM/SGLang inference engines. Fixes Fixed inability to enable continuous profiling over the Internet. Fixed Dify Plugin enhancement failures. Fixed incorrect upstream/downstream association in metrics. Fixed console redirection failures for certain database plugins. Fixed failures in pulling console configurations in some scenarios. Fixed duplicate request counting in certain Dify scenarios. Fixed failed application registration for Python 3.9 apps in some scenarios. Fixed request replay issues in urllib3 scenarios. Fixed inaccurate status code statistics in FastAPI. Fixed missing user_id and session_id collection in LangChain framework.
2.4.3	February 9, 2026	New features OpenAI SDK and Dashscope SDK now support multimodal data collection. For examples of output fields such as `gen_ai.input.messages` and `gen_ai.output.messages`, see the LLM section in the Attributes documentation. Added support for Claude Agent SDK framework. For more information, see Python components and frameworks supported by ARMS Application Monitoring. Fixes Fixed null pointer issues when using OpenTelemetry API in certain scenarios. Fixed package path conflicts when coexisting with other Alibaba Cloud SDKs. Fixed ineffective configuration updates when using older Protobuf versions. Fixed duplicate token recording in OpenAI and LangChain plugins under asynchronous scenarios. Fixed broken trace links between LangChain Chain nodes and other microservice call nodes. Fixed 405 status code responses caused by health check failures in certain scenarios.
2.3.5	January 16, 2026	New features Added continuous performance profiling for ongoing CPU and memory diagnostics in Python applications. For more information, see Continuous profiling for Python applications. Added database call metric collection for PyMySQL and psycopg2 frameworks. For more information, see Python components and frameworks supported by ARMS Application Monitoring. Added support for mem0 framework. For more information, see Python components and frameworks supported by ARMS Application Monitoring. Added instrumentation support for MCP framework versions ≤ 1.25.0. For more information, see Python components and frameworks supported by ARMS Application Monitoring. Added support for openai-agents framework. For more information, see Python components and frameworks supported by ARMS Application Monitoring. Fixes Fixed empty metric collection for gRPC streaming calls. Fixed incorrect error count metrics on the application overview page in certain scenarios. Fixed trace data reporting failures for Dify workflows caused by agent integration. Fixed potential performance degradation from divergent HTTP framework metric collection. Fixed potential performance degradation from divergent Redis framework metric collection. Fixed baggage setting failures via opentelemetry-api. Fixed missing data in OpenAI framework calls. Fixed misclassification of HTTP 400 status codes as 5xx errors in certain scenarios.
2.2.2	December 22, 2025	New features Added support for CrewAI framework. For more information, see Python components and frameworks supported by ARMS Application Monitoring. Added support for LiteLLM framework. For more information, see Python components and frameworks supported by ARMS Application Monitoring. Added support for AgentScope framework. For more information, see Python components and frameworks supported by ARMS Application Monitoring. Added support for compressing duplicate spans. For more information, see Integrating LLM applications/inference services with ARMS. Added compatibility with OpenTelemetry API v1.35.0 or later. For more information, see Python components and frameworks supported by ARMS Application Monitoring. Synchronized with the latest open source code from LoongSuite Python Agent. For details, see Changelog. Fixes Fixed missing first-packet latency metrics in LangChain plugin. Fixed missing tool names in LangChain plugin Tool operation metrics. Fixed missing first-packet latency metrics in OpenAI plugin. Fixed incorrect input/output and token statistics in Dashscope plugin streaming scenarios. Fixed missing Rerank operation metrics in Dashscope plugin.
2.1.3	December 9, 2025	Fixes Fixed LangChain and other plugins not working under Python 3.9. Fixed incorrect token statistics in LangChain plugin streaming scenarios. Fixed memory leaks in LangChain plugin in certain scenarios. Improved agent integration failures in multi-Python interpreter scenarios. Optimized detection logic for large model inference service frameworks. Fixed inaccurate agent instance count calculations.
2.1.2	November 24, 2025	New features Added support for Milvus vector database. For more information, see Python components and frameworks supported by ARMS Application Monitoring. Added observability data collection for Dify Plugin processes. For more information, see Python components and frameworks supported by ARMS Application Monitoring. Added support for BlackSheep framework. For more information, see Python components and frameworks supported by ARMS Application Monitoring. Added metric aggregation for Falcon, Starlette, gRPC, and MySQL frameworks. For more information, see Python components and frameworks supported by ARMS Application Monitoring. Added support for RagFlow framework. For more information, see Python components and frameworks supported by ARMS Application Monitoring. Added TraceID log correlation when using Loguru framework. For more information, see Python components and frameworks supported by ARMS Application Monitoring. Fixes Fixed duplicate token counting when using LangChain together with OpenAI/Dashscope frameworks. Fixed non-functional plugin toggle switches in the console. Fixed plugin enhancement failures under Python 3.8. Fixed negative GC count values in instance monitoring.
2.0.0	October 13, 2025	New features Added support for multiple new plugins including Google ADK, Tornado, and vLLM v1. For more information, see Python components and frameworks supported by ARMS Application Monitoring. Added model conversation history logging in application logs. For more information, see Controlling LLM application conversation history collection. Aligned OpenAI plugin input/output collection with the latest semantic specifications and added support for configuring maximum LLM content collection length. For more information, see LLM application semantic specification change history. Dify applications are now differentiated by workspace when splitting APPs. For more information, see Integrating LLM applications/inference services with ARMS. Fixes Optimized agent log output to avoid conflicts with user loggers. Fix the computational issue in the probe workspace.
1.8.3	September 19, 2025	New features Added support for limiting message content field length via environment variables. Fixes Unified OpenAI-related loggers to improve log management consistency. Fixed abnormal behavior when ARMS_WORKSPACE is empty.
1.8.2	September 12, 2025	New features Improved input/output display format for LLM spans in Langchain plugin. For more information, see LLM Trace field definitions. OpenAI plugin now supports metrics for LLM request count, error count, latency, and token consumption. For more information, see LLM metrics. OpenAI plugin now supports non-UTF-8 character inputs. Langchain plugin now includes a toggle for input/output content collection. For more information, see Integrating LLM applications/inference services with ARMS. Fixes Filtered out observability data from external calls related to ECS metadata to prevent interference with user queries. Fixed instance monitoring redirection failures for Python Agent in Cloud Monitor 2.0. Fixed span reporting failures when Python Agent is used alongside newer Langfuse SDK versions. Fixed error logs printed to stdout in non-cloud environments. Fixed aliyun package conflicts in non-intrusive integration scenarios. Fixed potential resource leaks during long-running agent sessions. Fixed slow startup in certain network-unavailable scenarios.
1.8.0	August 15, 2025	New features Added compatibility with newer opentelemetry-api versions (≤ 1.35.0). For more information, see Supported OpenTelemetry API versions. Added K8s/ECS metadata correlation to enhance cloud environment monitoring. Added distributed tracing support (propagation) for MCP protocol. Added support for MCP v1.12.4. For version support details, see Supported plugin versions. Added support for SGLang v0.4.9.post6. For version support details, see Supported plugin versions. LangChain now supports capturing user information such as sessionid and userid. OpenAI/LangChain now filters base64-encoded image data to avoid uploading large image payloads. Fixes Fixed memory leaks in Flask plugin under multi-process scenarios. Fixed errors and duplicate injections when Dify uses gevent. Fixed incompatibility with SGLang v0.4.6. Fixed unexpected None value errors during integration. Fixed compatibility issues with OpenAI plugin on Python versions below 3.11. Pinned cramjam version to 2.10.0 to resolve dependency conflicts under Python 3.8. Fixed broken traces in LangChain standalone mode. Standardized log levels for agent loading exceptions to improve troubleshooting.
1.6.1	July 28, 2025	Fixes Fixed unintended public network reporting of self-monitoring logs during health check failures.
1.6.0	July 22, 2025	New features Added support for SGlang large language model inference framework. For more information, see Observing vLLM/SGLang inference engines. Added support for MCP framework. For more information, see Python components and frameworks supported by ARMS Application Monitoring. Added support for Dify v1.4.0. For integration instructions, see Integrating Dify applications with ARMS Application Monitoring. Added support for specifying agent log path. For more information, see Common issues with Python Agent. Added support for specifying Python application type. For more information, see Integrating LLM applications/inference services with ARMS. Fixes Fixed inaccurate latency metrics in Flask plugin. Fixed inaccurate exception count metrics in Flask plugin. Fixed LLM application type detection failures. Fixed incorrect exception information display in openai plugin during call failures. Fixed occasional token calculation errors in LangChain plugin.
1.5.1	July 9, 2025	New features Dify plugin now supports the latest OTel specification. For details, see the pending specification PR. Added support for reporting data to Cloud Monitor 2.0. Fixes Fixed broken traces in Dify plugin Chat/Completion application calls. Fixed incompatibility errors when starting Dify services with gevent. Fixed broken upstream/downstream trace links in certain scenarios. Fixed package conflicts with other Alibaba Cloud third-party dependencies in certain scenarios.
1.4.0	June 23, 2025	New features Dify plugin now supports v1.3.0 and includes an input/output content collection toggle. Fixes Removed grpc-exporter dependencies to reduce potential conflicts. Fixed user call failures caused by DashScope streaming scenarios. Fixed attributes warnings caused by None values and disabled nacos error logs from stdout by default. Fixed missing spanKind tasks in Dify workflow spans. Fixed duplicate input/output issues in Dify plugin. Fixed unexpected responses during OpenAI asynchronous invocations.
1.3.1	May 23, 2025	Fixes Fixed occasional failures in retrieving service name and other metadata during agent data reporting. Fixed incorrect Python agent version display.
1.3.0	April 23, 2025	New features Added support for instance/application-level labels. Added distinction between LLM applications and LLM services. Fixes Limited Dify framework support range to prevent errors when users run unsupported Dify versions.
1.2.0	March 13, 2025	New features Added vLLM plugin. Langchain plugin now includes LLM metrics. Added support for Python 3.12. Fixes Resolved compatibility issues with OpenTelemetry API. Fixed Langchain plugin altering log levels. Fixed missing prompt information collection in Langchain plugin.
1.1.0	January 6, 2025	Added Dify plugin support. Added HTTPX plugin support. Added application business log correlation with trace IDs. For more information, see Correlating Python application business logs with trace IDs. OpenAI plugin now supports LLM application metrics. LlamaIndex plugin now supports LLM application metrics. LlamaIndex plugin now supports first-packet latency metrics. Fixed incorrect Flask metric labels. Fixed duplicate TraceId and SpanId when using Random operators. Fixed compatibility issues between agent and OpenTelemetry packages.
1.0.0	September 24, 2024	Enabled non-intrusive integration of Python applications with ARMS Application Monitoring, providing capabilities such as distributed tracing, performance monitoring, and intelligent alerting.