--- title: "Run Server Agent" method: POST path: "/agents/server/sessions/{session_id}/chat" tags: ["chat"] --- # Run Server Agent `POST /agents/server/sessions/{session_id}/chat` Stream a chat turn from the GraphQL server agent. This is the endpoint the PXI CLI talks to directly (no pre-configured agent record): it builds a fresh server agent per request from the caller-supplied model and contexts, then streams the reply back as Vercel-AI chunks. The request contexts gate capabilities — GraphQL mutations, web access, and subagents — and mutations are refused for viewer users. When trace recording is enabled (and permitted by system settings), the run is traced; locally ingested traces are persisted to the agent's project once the stream completes. Returns ``403`` if agents or the server agent are disabled, or if a viewer requests mutations. ## Path parameters - `session_id` string, required ## Request body - union — Discriminated union of chat request payloads. - ChatSubmitMessage — Submit message extended with Phoenix-specific fields. - `trigger` 'submit-message' - `id` string, required - `messages` PhoenixUIMessage[], required - `id` string, required - `role` 'system' | 'user' | 'assistant', required - `metadata` union - AssistantMessageMetadata — Wire schema for the chat stream's `message_metadata` payload. - `type` 'assistant' - `sessionId` string, required - `trace` AssistantMessageMetadataTraceIds - `traceId` string, required - `rootSpanId` string, required - `turnTraceContext` TurnTraceContext - `traceId` string, required - `rootSpanId` string, required - `startedAt` string, date-time, required - `usage` AssistantMessageMetadataUsage - `tokens` AssistantMessageMetadataUsageTokens, required - … - `promptDetails` AssistantMessageMetadataUsageTokenDetails - … - UserMessageMetadata — Wire schema for metadata the browser attaches to outgoing user messages. - `type` 'user' - `currentDateTime` string, required - `timeZone` string, required - `parts` union[], required - union - TextUIPart — A text part of a message. - `type` 'text' - `text` string, required - `state` 'streaming' | 'done', nullable - `providerMetadata` object, nullable - ReasoningUIPart — A reasoning part of a message. - `type` 'reasoning' - `text` string, required - `state` 'streaming' | 'done', nullable - `providerMetadata` object, nullable - ToolInputStreamingPart — Tool part in input-streaming state. - `type` string, required - `toolCallId` string, required - `title` string, nullable - `state` 'input-streaming' - `input` unknown - `providerExecuted` boolean, nullable - `callProviderMetadata` object, nullable - `approval` union - … - ToolInputAvailablePart — Tool part in input-available state. - `type` string, required - `toolCallId` string, required - `title` string, nullable - `state` 'input-available' - `input` unknown - `providerExecuted` boolean, nullable - `callProviderMetadata` object, nullable - `approval` union - … - ToolOutputAvailablePart — Tool part in output-available state. - `type` string, required - `toolCallId` string, required - `title` string, nullable - `state` 'output-available' - `input` unknown - `output` unknown - `providerExecuted` boolean, nullable - `callProviderMetadata` object, nullable - `preliminary` boolean, nullable - `approval` union - … - ToolOutputErrorPart — Tool part in output-error state. - `type` string, required - `toolCallId` string, required - `title` string, nullable - `state` 'output-error' - `input` unknown - `rawInput` unknown - `errorText` string, required - `providerExecuted` boolean, nullable - `callProviderMetadata` object, nullable - `approval` union - … - ToolApprovalRequestedPart — Tool part in approval-requested state (awaiting user decision). - `type` string, required - `toolCallId` string, required - `title` string, nullable - `state` 'approval-requested' - `input` unknown - `providerExecuted` boolean, nullable - `callProviderMetadata` object, nullable - `approval` union - … - ToolApprovalRespondedPart — Tool part in approval-responded state (user approved/denied, execution pending). - `type` string, required - `toolCallId` string, required - `title` string, nullable - `state` 'approval-responded' - `input` unknown - `providerExecuted` boolean, nullable - `callProviderMetadata` object, nullable - `approval` union - … - ToolOutputDeniedPart — Tool part in output-denied state (tool was denied, terminal state). - `type` string, required - `toolCallId` string, required - `title` string, nullable - `state` 'output-denied' - `input` unknown - `providerExecuted` boolean, nullable - `callProviderMetadata` object, nullable - `approval` union - … - DynamicToolInputStreamingPart — Dynamic tool part in input-streaming state. - `type` 'dynamic-tool' - `toolName` string, required - `toolCallId` string, required - `title` string, nullable - `state` 'input-streaming' - `input` unknown - `providerExecuted` boolean, nullable - `callProviderMetadata` object, nullable - `approval` union - … - DynamicToolInputAvailablePart — Dynamic tool part in input-available state. - `type` 'dynamic-tool' - `toolName` string, required - `toolCallId` string, required - `title` string, nullable - `state` 'input-available' - `input` unknown, required - `providerExecuted` boolean, nullable - `callProviderMetadata` object, nullable - `approval` union - … - DynamicToolOutputAvailablePart — Dynamic tool part in output-available state. - `type` 'dynamic-tool' - `toolName` string, required - `toolCallId` string, required - `title` string, nullable - `state` 'output-available' - `input` unknown, required - `output` unknown, required - `providerExecuted` boolean, nullable - `callProviderMetadata` object, nullable - `preliminary` boolean, nullable - `approval` union - … - DynamicToolOutputErrorPart — Dynamic tool part in output-error state. - `type` 'dynamic-tool' - `toolName` string, required - `toolCallId` string, required - `title` string, nullable - `state` 'output-error' - `input` unknown, required - `errorText` string, required - `providerExecuted` boolean, nullable - `callProviderMetadata` object, nullable - `approval` union - … - DynamicToolApprovalRequestedPart — Dynamic tool part in approval-requested state (awaiting user decision). - `type` 'dynamic-tool' - `toolName` string, required - `toolCallId` string, required - `title` string, nullable - `state` 'approval-requested' - `input` unknown, required - `providerExecuted` boolean, nullable - `callProviderMetadata` object, nullable - `approval` union - … - DynamicToolApprovalRespondedPart — Dynamic tool part in approval-responded state (user approved/denied, execution pending). - `type` 'dynamic-tool' - `toolName` string, required - `toolCallId` string, required - `title` string, nullable - `state` 'approval-responded' - `input` unknown, required - `providerExecuted` boolean, nullable - `callProviderMetadata` object, nullable - `approval` union - … - DynamicToolOutputDeniedPart — Dynamic tool part in output-denied state (tool was denied, terminal state). - `type` 'dynamic-tool' - `toolName` string, required - `toolCallId` string, required - `title` string, nullable - `state` 'output-denied' - `input` unknown, required - `providerExecuted` boolean, nullable - `callProviderMetadata` object, nullable - `approval` union - … - SourceUrlUIPart — A source part of a message. - `type` 'source-url' - `sourceId` string, required - `url` string, required - `title` string, nullable - `providerMetadata` object, nullable - SourceDocumentUIPart — A document source part of a message. - `type` 'source-document' - `sourceId` string, required - `mediaType` string, required - `title` string, required - `filename` string, nullable - `providerMetadata` object, nullable - FileUIPart — A file part of a message. - `type` 'file' - `mediaType` string, required - `filename` string, nullable - `url` string, required - `providerMetadata` object, nullable - DataUIPart — Data part with dynamic type based on data name. - `type` string, required - `id` string, nullable - `data` unknown, required - StepStartUIPart — A step boundary part of a message. - `type` 'step-start' - `ingestTraces` boolean - `exportRemoteTraces` boolean - `attachUserId` boolean — When true and the request is authenticated as a PhoenixUser, attaches the user's email as the OpenInference ``user.id`` span attribute on all traced work for this request. - `contexts` ChatContext[] - union — Discriminated union of every UI-state context the agent understands. - AppContext — Per-turn browser clock context for resolving relative time requests. - `type` 'app', required - `currentDateTime` string, required - `timeZone` string, required - ProjectContext — Project the user is currently viewing. ``span_filter`` carries the project-scoped span filter expression when the span filter field is mounted — empty string when the field is mounted with no condition applied, ``None`` when the field is not present at all. It describes the view in full, root-span scoping included (which is expressed within the filter DSL as ``parent_id is None``). - `type` 'project', required - `projectNodeId` string, required - `spanFilter` string, nullable - TraceContext - `type` 'trace', required - `projectNodeId` string, required - `otelTraceId` string, required - SessionContext — Session the user is currently viewing. - `type` 'session', required - `projectNodeId` string, required - `sessionNodeId` string, required - PromptContext — Prompt the user is currently viewing. - `type` 'prompt', required - `promptNodeId` string, required - PromptVersionContext — Prompt version the user is currently viewing. - `type` 'prompt_version', required - `promptNodeId` string, required - `promptVersionNodeId` string, required - AgentSpanContext — Span the user has selected. Exactly one of ``span_node_id`` (relay) or ``otel_span_id`` (OpenTelemetry hex) must be set. ``project_node_id`` is optional because a span can be selected from views outside a project route. - `type` 'span', required - `projectNodeId` string, nullable - `spanNodeId` string, nullable - `otelSpanId` string, nullable - PlaygroundContext — Playground prompt editor state mounted in the current browser route. - `type` 'playground', required - `recordExperiments` boolean - `repetitions` integer - `nextExperimentScaffold` PlaygroundExperimentScaffoldContext — Experiment name/description/metadata the user has staged for the playground's *next* dataset-backed run, before that run has started. The playground UI lets the user pre-set how the next recorded run's experiment will be named, described, and tagged (via the ``set_playground_experiment_recording`` tool or the recording form). That staged state is surfaced here so the agent can see what is already set and avoid re-staging it. Field semantics: - ``name`` / ``description``: the staged values, surfaced to the model verbatim, or ``None`` when the user has not staged them. - ``has_metadata``: a presence flag, not the value. Only *whether* metadata has been staged is model-relevant (so the agent knows not to re-attach it); the metadata object itself is deliberately kept out of the prompt. A field left unstaged (``None`` / ``False``) falls back to the server default when the run starts. The scaffold is consumed once that next run begins. - `name` string, nullable - `description` string, nullable - `hasMetadata` boolean - `instances` PlaygroundInstanceContext[] - `instanceId` integer, required - `model` union - … - `experimentId` string, nullable - `evaluators` PlaygroundEvaluatorContext[] - `datasetEvaluatorId` string, required - `name` string, required - `kind` 'LLM' | 'CODE' | 'BUILTIN', required - `isBuiltin` boolean, required - `isApplied` boolean, required - CodeEvaluatorContext — Code-evaluator create/edit form mounted in the current browser route. - `type` 'code_evaluator', required - `evaluatorNodeId` string, nullable - LlmEvaluatorContext — LLM-evaluator create/edit form mounted in the current browser route. - `type` 'llm_evaluator', required - `evaluatorNodeId` string, nullable - DatasetContext — Dataset the user is currently viewing or has bound to a workflow. Carries the dataset's relay node id and, when known, the active version node id. These IDs scope the create-form handoff link and the sampling of active dataset examples used as prompt context; the dataset schema itself is open. - `type` 'dataset', required - `datasetNodeId` string, required - `datasetVersionNodeId` string, nullable - GraphQLContext — GraphQL runtime state. - `type` 'graphql', required - `mutationsEnabled` boolean, required - WebAccessContext — User's per-turn request to expose web search / fetch tools. - `type` 'web_access', required - `enabled` boolean, required - SubagentsContext — User's per-turn request to expose the subagent-spawning tool. - `type` 'subagents', required - `enabled` boolean, required - `editPermission` 'manual' | 'bypass' - `requestedSkills` string[] — Skills the user explicitly requested via the prompt's slash-command affordance. The server force-loads each available skill by injecting a synthetic load_skill tool call/result at the tail of the message history. Unknown or context-unavailable names are ignored. - `model` union, required - CustomProviderModelSelection — Chat against a stored custom provider record. - `providerType` 'custom', required - `providerId` string, required - `modelName` string, required - BuiltInProviderModelSelection — Chat against a Phoenix built-in provider. Credentials and connection details (base URL, Azure endpoint, AWS region) are resolved from the secret store first and the process environment second. ``openai_api_type`` is honoured by the OpenAI and Azure OpenAI branches; other providers ignore it. - `providerType` 'builtin', required - `provider` 'OPENAI' | 'AZURE_OPENAI' | 'ANTHROPIC' | 'GOOGLE' | 'DEEPSEEK' | 'XAI' | 'OLLAMA' | 'AWS' | 'CEREBRAS' | 'FIREWORKS' | 'GROQ' | 'MOONSHOT' | 'PERPLEXITY' | 'TOGETHER', required - `modelName` string, required - `openaiApiType` 'chat_completions' | 'responses' - `turnTraceContext` TurnTraceContext - `traceId` string, required - `rootSpanId` string, required - `startedAt` string, date-time, required - ChatRegenerateMessage — Regenerate message extended with Phoenix-specific fields. - `trigger` 'regenerate-message', required - `id` string, required - `messages` PhoenixUIMessage[], required - `id` string, required - `role` 'system' | 'user' | 'assistant', required - `metadata` union - AssistantMessageMetadata — Wire schema for the chat stream's `message_metadata` payload. - `type` 'assistant' - `sessionId` string, required - `trace` AssistantMessageMetadataTraceIds - `traceId` string, required - `rootSpanId` string, required - `turnTraceContext` TurnTraceContext - `traceId` string, required - `rootSpanId` string, required - `startedAt` string, date-time, required - `usage` AssistantMessageMetadataUsage - `tokens` AssistantMessageMetadataUsageTokens, required - … - `promptDetails` AssistantMessageMetadataUsageTokenDetails - … - UserMessageMetadata — Wire schema for metadata the browser attaches to outgoing user messages. - `type` 'user' - `currentDateTime` string, required - `timeZone` string, required - `parts` union[], required - union - TextUIPart — A text part of a message. - `type` 'text' - `text` string, required - `state` 'streaming' | 'done', nullable - `providerMetadata` object, nullable - ReasoningUIPart — A reasoning part of a message. - `type` 'reasoning' - `text` string, required - `state` 'streaming' | 'done', nullable - `providerMetadata` object, nullable - ToolInputStreamingPart — Tool part in input-streaming state. - `type` string, required - `toolCallId` string, required - `title` string, nullable - `state` 'input-streaming' - `input` unknown - `providerExecuted` boolean, nullable - `callProviderMetadata` object, nullable - `approval` union - … - ToolInputAvailablePart — Tool part in input-available state. - `type` string, required - `toolCallId` string, required - `title` string, nullable - `state` 'input-available' - `input` unknown - `providerExecuted` boolean, nullable - `callProviderMetadata` object, nullable - `approval` union - … - ToolOutputAvailablePart — Tool part in output-available state. - `type` string, required - `toolCallId` string, required - `title` string, nullable - `state` 'output-available' - `input` unknown - `output` unknown - `providerExecuted` boolean, nullable - `callProviderMetadata` object, nullable - `preliminary` boolean, nullable - `approval` union - … - ToolOutputErrorPart — Tool part in output-error state. - `type` string, required - `toolCallId` string, required - `title` string, nullable - `state` 'output-error' - `input` unknown - `rawInput` unknown - `errorText` string, required - `providerExecuted` boolean, nullable - `callProviderMetadata` object, nullable - `approval` union - … - ToolApprovalRequestedPart — Tool part in approval-requested state (awaiting user decision). - `type` string, required - `toolCallId` string, required - `title` string, nullable - `state` 'approval-requested' - `input` unknown - `providerExecuted` boolean, nullable - `callProviderMetadata` object, nullable - `approval` union - … - ToolApprovalRespondedPart — Tool part in approval-responded state (user approved/denied, execution pending). - `type` string, required - `toolCallId` string, required - `title` string, nullable - `state` 'approval-responded' - `input` unknown - `providerExecuted` boolean, nullable - `callProviderMetadata` object, nullable - `approval` union - … - ToolOutputDeniedPart — Tool part in output-denied state (tool was denied, terminal state). - `type` string, required - `toolCallId` string, required - `title` string, nullable - `state` 'output-denied' - `input` unknown - `providerExecuted` boolean, nullable - `callProviderMetadata` object, nullable - `approval` union - … - DynamicToolInputStreamingPart — Dynamic tool part in input-streaming state. - `type` 'dynamic-tool' - `toolName` string, required - `toolCallId` string, required - `title` string, nullable - `state` 'input-streaming' - `input` unknown - `providerExecuted` boolean, nullable - `callProviderMetadata` object, nullable - `approval` union - … - DynamicToolInputAvailablePart — Dynamic tool part in input-available state. - `type` 'dynamic-tool' - `toolName` string, required - `toolCallId` string, required - `title` string, nullable - `state` 'input-available' - `input` unknown, required - `providerExecuted` boolean, nullable - `callProviderMetadata` object, nullable - `approval` union - … - DynamicToolOutputAvailablePart — Dynamic tool part in output-available state. - `type` 'dynamic-tool' - `toolName` string, required - `toolCallId` string, required - `title` string, nullable - `state` 'output-available' - `input` unknown, required - `output` unknown, required - `providerExecuted` boolean, nullable - `callProviderMetadata` object, nullable - `preliminary` boolean, nullable - `approval` union - … - DynamicToolOutputErrorPart — Dynamic tool part in output-error state. - `type` 'dynamic-tool' - `toolName` string, required - `toolCallId` string, required - `title` string, nullable - `state` 'output-error' - `input` unknown, required - `errorText` string, required - `providerExecuted` boolean, nullable - `callProviderMetadata` object, nullable - `approval` union - … - DynamicToolApprovalRequestedPart — Dynamic tool part in approval-requested state (awaiting user decision). - `type` 'dynamic-tool' - `toolName` string, required - `toolCallId` string, required - `title` string, nullable - `state` 'approval-requested' - `input` unknown, required - `providerExecuted` boolean, nullable - `callProviderMetadata` object, nullable - `approval` union - … - DynamicToolApprovalRespondedPart — Dynamic tool part in approval-responded state (user approved/denied, execution pending). - `type` 'dynamic-tool' - `toolName` string, required - `toolCallId` string, required - `title` string, nullable - `state` 'approval-responded' - `input` unknown, required - `providerExecuted` boolean, nullable - `callProviderMetadata` object, nullable - `approval` union - … - DynamicToolOutputDeniedPart — Dynamic tool part in output-denied state (tool was denied, terminal state). - `type` 'dynamic-tool' - `toolName` string, required - `toolCallId` string, required - `title` string, nullable - `state` 'output-denied' - `input` unknown, required - `providerExecuted` boolean, nullable - `callProviderMetadata` object, nullable - `approval` union - … - SourceUrlUIPart — A source part of a message. - `type` 'source-url' - `sourceId` string, required - `url` string, required - `title` string, nullable - `providerMetadata` object, nullable - SourceDocumentUIPart — A document source part of a message. - `type` 'source-document' - `sourceId` string, required - `mediaType` string, required - `title` string, required - `filename` string, nullable - `providerMetadata` object, nullable - FileUIPart — A file part of a message. - `type` 'file' - `mediaType` string, required - `filename` string, nullable - `url` string, required - `providerMetadata` object, nullable - DataUIPart — Data part with dynamic type based on data name. - `type` string, required - `id` string, nullable - `data` unknown, required - StepStartUIPart — A step boundary part of a message. - `type` 'step-start' - `messageId` string, nullable - `ingestTraces` boolean - `exportRemoteTraces` boolean - `attachUserId` boolean — When true and the request is authenticated as a PhoenixUser, attaches the user's email as the OpenInference ``user.id`` span attribute on all traced work for this request. - `contexts` ChatContext[] - union — Discriminated union of every UI-state context the agent understands. - AppContext — Per-turn browser clock context for resolving relative time requests. - `type` 'app', required - `currentDateTime` string, required - `timeZone` string, required - ProjectContext — Project the user is currently viewing. ``span_filter`` carries the project-scoped span filter expression when the span filter field is mounted — empty string when the field is mounted with no condition applied, ``None`` when the field is not present at all. It describes the view in full, root-span scoping included (which is expressed within the filter DSL as ``parent_id is None``). - `type` 'project', required - `projectNodeId` string, required - `spanFilter` string, nullable - TraceContext - `type` 'trace', required - `projectNodeId` string, required - `otelTraceId` string, required - SessionContext — Session the user is currently viewing. - `type` 'session', required - `projectNodeId` string, required - `sessionNodeId` string, required - PromptContext — Prompt the user is currently viewing. - `type` 'prompt', required - `promptNodeId` string, required - PromptVersionContext — Prompt version the user is currently viewing. - `type` 'prompt_version', required - `promptNodeId` string, required - `promptVersionNodeId` string, required - AgentSpanContext — Span the user has selected. Exactly one of ``span_node_id`` (relay) or ``otel_span_id`` (OpenTelemetry hex) must be set. ``project_node_id`` is optional because a span can be selected from views outside a project route. - `type` 'span', required - `projectNodeId` string, nullable - `spanNodeId` string, nullable - `otelSpanId` string, nullable - PlaygroundContext — Playground prompt editor state mounted in the current browser route. - `type` 'playground', required - `recordExperiments` boolean - `repetitions` integer - `nextExperimentScaffold` PlaygroundExperimentScaffoldContext — Experiment name/description/metadata the user has staged for the playground's *next* dataset-backed run, before that run has started. The playground UI lets the user pre-set how the next recorded run's experiment will be named, described, and tagged (via the ``set_playground_experiment_recording`` tool or the recording form). That staged state is surfaced here so the agent can see what is already set and avoid re-staging it. Field semantics: - ``name`` / ``description``: the staged values, surfaced to the model verbatim, or ``None`` when the user has not staged them. - ``has_metadata``: a presence flag, not the value. Only *whether* metadata has been staged is model-relevant (so the agent knows not to re-attach it); the metadata object itself is deliberately kept out of the prompt. A field left unstaged (``None`` / ``False``) falls back to the server default when the run starts. The scaffold is consumed once that next run begins. - `name` string, nullable - `description` string, nullable - `hasMetadata` boolean - `instances` PlaygroundInstanceContext[] - `instanceId` integer, required - `model` union - … - `experimentId` string, nullable - `evaluators` PlaygroundEvaluatorContext[] - `datasetEvaluatorId` string, required - `name` string, required - `kind` 'LLM' | 'CODE' | 'BUILTIN', required - `isBuiltin` boolean, required - `isApplied` boolean, required - CodeEvaluatorContext — Code-evaluator create/edit form mounted in the current browser route. - `type` 'code_evaluator', required - `evaluatorNodeId` string, nullable - LlmEvaluatorContext — LLM-evaluator create/edit form mounted in the current browser route. - `type` 'llm_evaluator', required - `evaluatorNodeId` string, nullable - DatasetContext — Dataset the user is currently viewing or has bound to a workflow. Carries the dataset's relay node id and, when known, the active version node id. These IDs scope the create-form handoff link and the sampling of active dataset examples used as prompt context; the dataset schema itself is open. - `type` 'dataset', required - `datasetNodeId` string, required - `datasetVersionNodeId` string, nullable - GraphQLContext — GraphQL runtime state. - `type` 'graphql', required - `mutationsEnabled` boolean, required - WebAccessContext — User's per-turn request to expose web search / fetch tools. - `type` 'web_access', required - `enabled` boolean, required - SubagentsContext — User's per-turn request to expose the subagent-spawning tool. - `type` 'subagents', required - `enabled` boolean, required - `editPermission` 'manual' | 'bypass' - `requestedSkills` string[] — Skills the user explicitly requested via the prompt's slash-command affordance. The server force-loads each available skill by injecting a synthetic load_skill tool call/result at the tail of the message history. Unknown or context-unavailable names are ignored. - `model` union, required - CustomProviderModelSelection — Chat against a stored custom provider record. - `providerType` 'custom', required - `providerId` string, required - `modelName` string, required - BuiltInProviderModelSelection — Chat against a Phoenix built-in provider. Credentials and connection details (base URL, Azure endpoint, AWS region) are resolved from the secret store first and the process environment second. ``openai_api_type`` is honoured by the OpenAI and Azure OpenAI branches; other providers ignore it. - `providerType` 'builtin', required - `provider` 'OPENAI' | 'AZURE_OPENAI' | 'ANTHROPIC' | 'GOOGLE' | 'DEEPSEEK' | 'XAI' | 'OLLAMA' | 'AWS' | 'CEREBRAS' | 'FIREWORKS' | 'GROQ' | 'MOONSHOT' | 'PERPLEXITY' | 'TOGETHER', required - `modelName` string, required - `openaiApiType` 'chat_completions' | 'responses' - `turnTraceContext` TurnTraceContext - `traceId` string, required - `rootSpanId` string, required - `startedAt` string, date-time, required ## Response `200` Successful Response - unknown ## Other responses - `422` — Validation Error --- [API](https://skmtc.net/arize-ai/apis/arize-phoenix-rest-api.md) · [All operations](https://skmtc.net/arize-ai/apis/arize-phoenix-rest-api/llms.txt) · [OpenAPI document](https://skmtc-service-staging.skmtc.workers.dev/v1/apis/arize-ai/arize-phoenix-rest-api/versions/a14d8ad6f708/schema)