Arize-ai/phoenix

Phoenix 是一个开源的 AI 可观测性平台，专为实验、评估和故障排除而设计。它提供： - [**_Tracing_**](https://arize.com/docs/phoenix/tracing/llm-traces) - 使用基于 OpenTelemetry 的埋点追踪你的 LLM 应用程序的 runtime。 - [**_Evaluation_**](https://arize.com/docs/phoenix/evaluation/llm-evals) - 利用 LLM 通过响应和检索评估来对你的应用程序性能进行基准测试。 - [**_Datasets_**](https://arize.com/docs/phoenix/datasets-and-experiments/overview-datasets) - 创建带有版本控制的示例数据集，用于实验、评估和 fine-tuning。 - [**_Experiments_**](https://arize.com/docs/phoenix/datasets-and-experiments/overview-datasets#experiments) - 跟踪并评估对 prompts、LLM 和检索的更改。 - [**_Playground_**](https://arize.com/docs/phoenix/prompt-engineering/overview-prompts)- 优化 prompts，比较模型，调整参数，并重放已追踪的 LLM 调用。 - [**_Prompt Management_**](https://arize.com/docs/phoenix/prompt-engineering/overview-prompts/prompt-management)- 使用版本控制、标签和实验来系统化地管理和测试 prompt 更改。 - [**_PXI (Phoenix Intelligence)_**](https://arize.com/docs/phoenix/pxi) - 内置于 Phoenix 中的 AI 工程代理，用于调试 traces、迭代 prompts 以及导航产品。 Phoenix 与供应商和语言无关，开箱即支持主流的框架（[OpenAI Agents SDK](https://arize.com/docs/phoenix/tracing/integrations-tracing/openai-agents-sdk), [Claude Agent SDK](https://arize.com/docs/phoenix/integrations/python/claude-agent-sdk), [LangGraph](https://arize.com/docs/phoenix/tracing/integrations-tracing/langchain), [Vercel AI SDK](https://arize.com/docs/phoenix/tracing/integrations-tracing/vercel-ai-sdk), [Mastra](https://arize.com/docs/phoenix/integrations/typescript/mastra), [CrewAI](https://arize.com/docs/phoenix/tracing/integrations-tracing/crewai), [LlamaIndex](https://arize.com/docs/phoenix/tracing/integrations-tracing/llamaindex), [DSPy](https://arize.com/docs/phoenix/tracing/integrations-tracing/dspy)）和 LLM 提供商（[OpenAI](https://arize.com/docs/phoenix/tracing/integrations-tracing/openai), [Anthropic](https://arize.com/docs/phoenix/tracing/integrations-tracing/anthropic), [Google GenAI](https://arize.com/docs/phoenix/tracing/integrations-tracing/google-genai), [Google ADK](https://arize.com/docs/phoenix/integrations/llm-providers/google-gen-ai/google-adk-tracing), [AWS Bedrock](https://arize.com/docs/phoenix/tracing/integrations-tracing/bedrock), [OpenRouter](https://arize.com/docs/phoenix/integrations/python/openrouter), [LiteLLM](https://arize.com/docs/phoenix/tracing/integrations-tracing/litellm) 等）。有关自动埋点的详细信息，请查看 [OpenInference](https://github.com/Arize-ai/openinference) 项目。 Phoenix 几乎可以在任何地方运行，包括你的本地机器、Jupyter notebook、容器化部署或云端。 ## 安装说明 通过 `pip` 或 `conda` 安装 Phoenix ``` pip install arize-phoenix ``` Phoenix 容器镜像可通过 [Docker Hub](https://hub.docker.com/r/arizephoenix/phoenix) 获取，并可以使用 Docker 或 Kubernetes 进行部署。Arize AI 还在 [app.phoenix.arize.com](https://app.phoenix.arize.com/) 提供了云实例。 ## 包 `arize-phoenix` 包包含了整个 Phoenix 平台。但是，如果你已经部署了 Phoenix 平台，则可以使用轻量级的 Python 子包和 TypeScript 包与该平台配合使用。 ### Python 子包 | Package | Version & Docs | Description | | --------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------ | | [arize-phoenix-otel](https://github.com/Arize-ai/phoenix/tree/main/packages/phoenix-otel) | [](https://pypi.org/project/arize-phoenix-otel/) [](https://arize-phoenix.readthedocs.io/projects/otel/en/latest/index.html) | 为 OpenTelemetry 原语提供了一个带有 Phoenix 感知默认值的轻量级封装 | | [arize-phoenix-client](https://github.com/Arize-ai/phoenix/tree/main/packages/phoenix-client) | [](https://pypi.org/project/arize-phoenix-client/) [](https://arize-phoenix.readthedocs.io/projects/client/en/latest/index.html) | 通过其 OpenAPI REST 接口与 Phoenix server 交互的轻量级 client | | [arize-phoenix-evals](https://github.com/Arize-ai/phoenix/tree/main/packages/phoenix-evals) | [](https://pypi.org/project/arize-phoenix-evals/) [](https://arize-phoenix.readthedocs.io/projects/evals/en/latest/index.html) | 用于评估 LLM 应用程序的工具，包括 RAG 相关性、回答相关性等 | ### TypeScript 子包 | Package | Version & Docs | Description | | --------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------ | | [@arizeai/phoenix-otel](https://github.com/Arize-ai/phoenix/tree/main/js/packages/phoenix-otel) | [](https://www.npmjs.com/package/@arizeai/phoenix-otel) [](https://arize-ai.github.io/phoenix/) | 为 OpenTelemetry 原语提供了一个带有 Phoenix 感知默认值的轻量级封装 | | [@arizeai/phoenix-client](https://github.com/Arize-ai/phoenix/tree/main/js/packages/phoenix-client) | [](https://www.npmjs.com/package/@arizeai/phoenix-client) [](https://arize-ai.github.io/phoenix/) | 用于 Arize Phoenix API 的 client | | [@arizeai/phoenix-evals](https://github.com/Arize-ai/phoenix/tree/main/js/packages/phoenix-evals) | [](https://www.npmjs.com/package/@arizeai/phoenix-evals) [](https://arize-ai.github.io/phoenix/) | 用于 LLM 应用程序的 TypeScript 评估库 (alpha 版本) | | [@arizeai/phoenix-mcp](https://github.com/Arize-ai/phoenix/tree/main/js/packages/phoenix-mcp) | [](https://www.npmjs.com/package/@arizeai/phoenix-mcp) [](./js/packages/phoenix-mcp/README.md) | Arize Phoenix 的 MCP server 实现，提供访问 Phoenix 功能的统一接口 | | [@arizeai/phoenix-cli](https://github.com/Arize-ai/phoenix/tree/main/js/packages/phoenix-cli) | [](https://www.npmjs.com/package/@arizeai/phoenix-cli) [](https://arize.com/docs/phoenix/sdk-api-reference/typescript/arizeai-phoenix-cli) | 用于获取 traces、数据集和实验的 CLI，可与 Claude Code、Cursor 及其他编程代理配合使用 | ## Tracing 集成 Phoenix 基于 OpenTelemetry 构建，与供应商、语言和框架无关。有关 tracing 集成和应用程序示例的详细信息，请参阅 [OpenInference](https://github.com/Arize-ai/openinference) 项目。 **Python 集成** | | Integration | Package | Version | |:---:|---|---|---| | | [OpenAI](https://arize.com/docs/phoenix/tracing/integrations-tracing/openai) | `openinference-instrumentation-openai` | [](https://pypi.python.org/pypi/openinference-instrumentation-openai) | | | [OpenAI Agents](https://arize.com/docs/phoenix/tracing/integrations-tracing/openai-agents-sdk) | `openinference-instrumentation-openai-agents` | [](https://pypi.python.org/pypi/openinference-instrumentation-openai-agents) | | | [LlamaIndex](https://arize.com/docs/phoenix/tracing/integrations-tracing/llamaindex) | `openinference-instrumentation-llama-index` | [](https://pypi.python.org/pypi/openinference-instrumentation-llama-index) | | | [DSPy](https://arize.com/docs/phoenix/tracing/integrations-tracing/dspy) | `openinference-instrumentation-dspy` | [](https://pypi.python.org/pypi/openinference-instrumentation-dspy) | | | [AWS Bedrock](https://arize.com/docs/phoenix/tracing/integrations-tracing/bedrock) | `openinference-instrumentation-bedrock` | [](https://pypi.python.org/pypi/openinference-instrumentation-bedrock) | | | [LangChain](https://arize.com/docs/phoenix/tracing/integrations-tracing/langchain) | `openinference-instrumentation-langchain` | [](https://pypi.python.org/pypi/openinference-instrumentation-langchain) | | | [MistralAI](https://arize.com/docs/phoenix/tracing/integrations-tracing/mistralai) | `openinference-instrumentation-mistralai` | [](https://pypi.python.org/pypi/openinference-instrumentation-mistralai) | | | [Google GenAI](https://arize.com/docs/phoenix/tracing/integrations-tracing/google-gen-ai) | `openinference-instrumentation-google-genai` | [](https://pypi.python.org/pypi/openinference-instrumentation-google-genai) | | | [Google ADK](https://arize.com/docs/phoenix/integrations/llm-providers/google-gen-ai/google-adk-tracing) | `openinference-instrumentation-google-adk` | [](https://pypi.python.org/pypi/openinference-instrumentation-google-adk) | | | [Guardrails](https://arize.com/docs/phoenix/tracing/integrations-tracing/guardrails) | `openinference-instrumentation-guardrails` | [](https://pypi.python.org/pypi/openinference-instrumentation-guardrails) | | | [VertexAI](https://arize.com/docs/phoenix/tracing/integrations-tracing/vertexai) | `openinference-instrumentation-vertexai` | [](https://pypi.python.org/pypi/openinference-instrumentation-vertexai) | | | [CrewAI](https://arize.com/docs/phoenix/tracing/integrations-tracing/crewai) | `openinference-instrumentation-crewai` | [](https://pypi.python.org/pypi/openinference-instrumentation-crewai) | | | [Haystack](https://arize.com/docs/phoenix/tracing/integrations-tracing/haystack) | `openinference-instrumentation-haystack` | [](https://pypi.python.org/pypi/openinference-instrumentation-haystack) | | | [LiteLLM](https://arize.com/docs/phoenix/tracing/integrations-tracing/litellm) | `openinference-instrumentation-litellm` | [](https://pypi.python.org/pypi/openinference-instrumentation-litellm) | | | [Groq](https://arize.com/docs/phoenix/tracing/integrations-tracing/groq) | `openinference-instrumentation-groq` | [](https://pypi.python.org/pypi/openinference-instrumentation-groq) | | | [Instructor](https://arize.com/docs/phoenix/tracing/integrations-tracing/instructor) | `openinference-instrumentation-instructor` | [](https://pypi.python.org/pypi/openinference-instrumentation-instructor) | | | [Anthropic](https://arize.com/docs/phoenix/tracing/integrations-tracing/anthropic) | `openinference-instrumentation-anthropic` | [](https://pypi.python.org/pypi/openinference-instrumentation-anthropic) | | | [Smolagents](https://huggingface.co/docs/smolagents/en/tutorials/inspect_runs) | `openinference-instrumentation-smolagents` | [](https://pypi.python.org/pypi/openinference-instrumentation-smolagents) | | | [Agno](https://arize.com/docs/phoenix/tracing/integrations-tracing/agno) | `openinference-instrumentation-agno` | [](https://pypi.python.org/pypi/openinference-instrumentation-agno) | | | [MCP](https://arize.com/docs/phoenix/tracing/integrations-tracing/model-context-protocol-mcp) | `openinference-instrumentation-mcp` | [](https://pypi.python.org/pypi/openinference-instrumentation-mcp) | | | [Pydantic AI](https://arize.com/docs/phoenix/integrations/python/pydantic) | `openinference-instrumentation-pydantic-ai` | [](https://pypi.python.org/pypi/openinference-instrumentation-pydantic-ai) | | | [Autogen AgentChat](https://arize.com/docs/phoenix/integrations/frameworks/autogen/autogen-tracing) | `openinference-instrumentation-autogen-agentchat` | [](https://pypi.python.org/pypi/openinference-instrumentation-autogen-agentchat) | | | [Portkey](https://arize.com/docs/phoenix/integrations/portkey) | `openinference-instrumentation-portkey` | [](https://pypi.python.org/pypi/openinference-instrumentation-portkey) | | | [Agent Spec](https://arize.com/docs/phoenix/tracing/integrations-tracing/agentspec) | `openinference-instrumentation-agentspec` | [](https://pypi.python.org/pypi/openinference-instrumentation-agentspec) | | | [Claude Agent SDK](https://arize.com/docs/phoenix/integrations/python/claude-agent-sdk) | `openinference-instrumentation-claude-agent-sdk` | [](https://pypi.python.org/pypi/openinference-instrumentation-claude-agent-sdk) | ## Span Processors 通过添加统一数据的 span processors，跨其他埋点库规范化和转换数据。 | Package | Description | Version | | ----------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | [`openinference-instrumentation-openlit`](./python/instrumentation/openinference-instrumentation-openlit) | OpenInference Span Processor for OpenLIT traces. | [](https://pypi.python.org/pypi/openinference-instrumentation-openlit) | | [`openinference-instrumentation-openllmetry`](./python/instrumentation/openinference-instrumentation-openllmetry) | OpenInference Span Processor for OpenLLMetry (Traceloop) traces. | [](https://pypi.python.org/pypi/openinference-instrumentation-openllmetry) | ### JavaScript 集成 | | Integration | Package | Version | |:---:|---|---|---| | | [OpenAI](https://arize.com/docs/phoenix/tracing/integrations-tracing/openai-node-sdk) | `@arizeai/openinference-instrumentation-openai` | [](https://www.npmjs.com/package/@arizeai/openinference-instrumentation-openai) | | | [OpenAI Agents](https://arize.com/docs/phoenix/integrations/typescript/openai-agents) | `@arizeai/openinference-instrumentation-openai-agents` | [](https://www.npmjs.com/package/@arizeai/openinference-instrumentation-openai-agents) | | | [LangChain.js](https://arize.com/docs/phoenix/tracing/integrations-tracing/langchain) | `@arizeai/openinference-instrumentation-langchain` | [](https://www.npmjs.com/package/@arizeai/openinference-instrumentation-langchain) | | | [Vercel AI SDK](https://arize.com/docs/phoenix/tracing/integrations-tracing/vercel-ai-sdk) | `@arizeai/openinference-vercel` | [](https://www.npmjs.com/package/@arizeai/openinference-vercel) | | | [BeeAI](https://arize.com/docs/phoenix/tracing/integrations-tracing/beeai) | `@arizeai/openinference-instrumentation-beeai` | [](https://www.npmjs.com/package/@arizeai/openinference-instrumentation-beeai) | | | [Claude Agent SDK](https://arize.com/docs/phoenix/integrations/typescript/claude-agent-sdk) | `@arizeai/openinference-instrumentation-claude-agent-sdk` | [](https://www.npmjs.com/package/@arizeai/openinference-instrumentation-claude-agent-sdk) | | | [Mastra](https://arize.com/docs/phoenix/integrations/typescript/mastra) | `@mastra/arize` | [](https://www.npmjs.com/package/@mastra/arize) | | | [MCP](https://arize.com/docs/phoenix/integrations/typescript/mcp) | `@arizeai/openinference-instrumentation-mcp` | [](https://www.npmjs.com/package/@arizeai/openinference-instrumentation-mcp) | ### Java 集成 | | Integration | Package | Version | |:---:|---|---|---| | | [LangChain4j](https://github.com/Arize-ai/openinference/tree/main/java/instrumentation/openinference-instrumentation-langchain4j) | `openinference-instrumentation-langchain4j` | [](https://central.sonatype.com/artifact/com.arize/openinference-instrumentation-langchain4j) | | | SpringAI | `openinference-instrumentation-springAI` | [](https://central.sonatype.com/artifact/com.arize/openinference-instrumentation-springAI) | | | [Arconia](https://docs.arconia.io/arconia/latest/observability/semantic-conventions/openinference/) for Spring AI | `io.arconia:arconia-openinference-semantic-conventions` | [](https://central.sonatype.com/artifact/io.arconia/arconia-openinference-ai-semantic-conventions) | ### Go 集成 | | Integration | Package | Version | |:---:|---|---|---| | | [OpenAI](https://github.com/Arize-ai/openinference/tree/main/go/openinference-instrumentation-openai-go) | `github.com/Arize-ai/openinference/go/openinference-instrumentation-openai-go` | [](https://pkg.go.dev/github.com/Arize-ai/openinference/go/openinference-instrumentation-openai-go) | | | [Anthropic](https://github.com/Arize-ai/openinference/tree/main/go/openinference-instrumentation-anthropic-sdk-go) | `github.com/Arize-ai/openinference/go/openinference-instrumentation-anthropic-sdk-go` | [](https://pkg.go.dev/github.com/Arize-ai/openinference/go/openinference-instrumentation-anthropic-sdk-go) | ### 平台 | | Platform | Description | Docs | |:---:|---|---|---| | | [BeeAI](https://docs.beeai.dev/observability/agents-traceability) | 带有内置可观测性的 AI 代理框架 | [集成指南](https://docs.beeai.dev/observability/agents-traceability) | | | [Dify](https://docs.dify.ai/en/guides/monitoring/integrate-external-ops-tools/integrate-phoenix) | 开源 LLM 应用程序开发平台 | [集成指南](https://docs.dify.ai/en/guides/monitoring/integrate-external-ops-tools/integrate-phoenix) | | | [Envoy AI Gateway](https://github.com/envoyproxy/ai-gateway) | 基于 Envoy Proxy 构建的、用于 AI 工作负载的 AI Gateway | [集成指南](https://github.com/envoyproxy/ai-gateway/tree/main/cmd/aigw#opentelemetry-setup-with-phoenix) | | | [LangFlow](https://arize.com/docs/phoenix/tracing/integrations-tracing/langflow) | 用于构建多代理和 RAG 应用程序的可视化框架 | [集成指南](https://arize.com/docs/phoenix/tracing/integrations-tracing/langflow) | | | [LiteLLM Proxy](https://docs.litellm.ai/docs/observability/phoenix_integration#using-with-litellm-proxy) | 用于 LLM 的代理服务器 | [集成指南](https://docs.litellm.ai/docs/observability/phoenix_integration#using-with-litellm-proxy) | | | [Flowise](https://arize.com/docs/phoenix/integrations/platforms/flowise) | 用于构建 LLM 应用程序的可视化框架 | [集成指南](https://arize.com/docs/phoenix/integrations/platforms/flowise) | | | [Prompt Flow](https://arize.com/docs/phoenix/integrations/platforms/prompt-flow) | Microsoft 的 prompt flow 编排工具 | [集成指南](https://arize.com/docs/phoenix/integrations/platforms/prompt-flow) | | | [NVIDIA NeMo](https://arize.com/docs/phoenix/integrations/python/nvidia) | 用于企业级代理的 NVIDIA NeMo Agent Toolkit | [集成指南](https://arize.com/docs/phoenix/integrations/python/nvidia) | | | [Graphite](https://arize.com/docs/phoenix/integrations/python/graphite) | 带有可视化构建器的多代理 LLM 工作流框架 | [集成指南](https://arize.com/docs/phoenix/integrations/python/graphite) | ## 编程代理技能 此代码仓库包含[技能](https://docs.anthropic.com/en/docs/claude-code/skills)，可教导编程代理如何使用 Phoenix。它们位于 [`.agents/skills/`](.agents/skills/) 中，可与 Claude Code、Cursor 及其他兼容工具一起使用。 | Skill | Description | | ----- | ----------- | | [phoenix-cli](.agents/skills/phoenix-cli/) | 使用 Phoenix CLI 调试 LLM 应用程序 — 获取 traces、分析错误、查看实验并查询 API | | [phoenix-evals](.agents/skills/phoenix-evals/) | 使用 Phoenix 构建并运行用于 AI/LLM 应用程序的评估器 | | [phoenix-tracing](.agents/skills/phoenix-tracing/) | 用于追踪 LLM 应用程序的 OpenInference 语义约定和埋点 | ## 安全与隐私 我们非常重视数据安全和隐私。有关更多详细信息，请参阅我们的[安全与隐私文档](https://arize.com/docs/phoenix/self-hosting/security/privacy)。 ### 遥测 默认情况下，Phoenix 会收集基本的网络分析数据（例如，页面浏览量、UI 交互），以帮助我们了解 Phoenix 的使用情况并改进产品。**我们绝不会收集你的任何 trace 数据、评估结果或任何敏感信息。** 你可以通过设置环境变量来选择退出遥测：`PHOENIX_TELEMETRY_ENABLED=false` ## 破坏性更改 有关破坏性更改的列表，请参阅[迁移指南](./MIGRATION.md)。 ## 版权、专利与许可 版权所有 2025 Arize AI, Inc. 保留所有权利。 此代码的部分内容受一项或多项美国专利保护。请参阅 [IP_NOTICE](https://github.com/Arize-ai/phoenix/blob/main/IP_NOTICE)。 此软件根据 Elastic License 2.0 (ELv2) 的条款获得许可。请参阅 [LICENSE](https://github.com/Arize-ai/phoenix/blob/main/LICENSE)。

标签：API集成, DLL 劫持, SOC Prime, 人工智能, 可观测性, 大语言模型, 子域名突变, 开发工具, 用户代理, 用户模式Hook绕过, 评估平台, 请求拦截, 逆向工具