@arizeai/phoenix-client 6.10.1 → 6.11.1

package/src/__generated__/api/v1.ts

@@ -388,7 +388,11 @@ export interface paths {
         delete: operations["deleteExperiment"];
         options?: never;
         head?: never;
-        patch?: never;
+        /**
+         * Update an experiment by ID
+         * @description Partially update an experiment's name, description, and/or metadata. Only the fields included in the request body are changed; omitted fields are left as-is. Patching an ephemeral experiment refreshes its last-update timestamp, which extends the window before it is swept away.
+         */
+        patch: operations["updateExperiment"];
         trace?: never;
     };
     "/v1/experiments/{experiment_id}/incomplete-runs": {
@@ -1294,6 +1298,23 @@ export interface paths {
         patch?: never;
         trace?: never;
     };
+    "/agents/server/sessions/{session_id}/chat": {
+        parameters: {
+            query?: never;
+            header?: never;
+            path?: never;
+            cookie?: never;
+        };
+        get?: never;
+        put?: never;
+        /** Run Server Agent */
+        post: operations["run_server_agent_agents_server_sessions__session_id__chat_post"];
+        delete?: never;
+        options?: never;
+        head?: never;
+        patch?: never;
+        trace?: never;
+    };
     "/agents/{agent_id}/sessions/{session_id}/chat": {
         parameters: {
             query?: never;
@@ -1610,7 +1631,7 @@ export interface components {
          * ChatContext
          * @description Discriminated union of every UI-state context the agent understands.
          */
-        ChatContext: components["schemas"]["AppContext"] | components["schemas"]["ProjectContext"] | components["schemas"]["TraceContext"] | components["schemas"]["AgentSpanContext"] | components["schemas"]["PlaygroundContext"] | components["schemas"]["CodeEvaluatorContext"] | components["schemas"]["LlmEvaluatorContext"] | components["schemas"]["DatasetContext"] | components["schemas"]["GraphQLContext"] | components["schemas"]["WebAccessContext"];
+        ChatContext: components["schemas"]["AppContext"] | components["schemas"]["ProjectContext"] | components["schemas"]["TraceContext"] | components["schemas"]["SessionContext"] | components["schemas"]["PromptContext"] | components["schemas"]["PromptVersionContext"] | components["schemas"]["AgentSpanContext"] | components["schemas"]["PlaygroundContext"] | components["schemas"]["CodeEvaluatorContext"] | components["schemas"]["LlmEvaluatorContext"] | components["schemas"]["DatasetContext"] | components["schemas"]["GraphQLContext"] | components["schemas"]["WebAccessContext"] | components["schemas"]["SubagentsContext"];
         /**
          * ChatRegenerateMessage
          * @description Regenerate message extended with Phoenix-specific fields.
@@ -1637,6 +1658,12 @@ export interface components {
              * @default false
              */
             exportRemoteTraces?: boolean;
+            /**
+             * Attachuserid
+             * @description 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.
+             * @default false
+             */
+            attachUserId?: boolean;
             /** Contexts */
             contexts?: components["schemas"]["ChatContext"][];
             /**
@@ -1684,6 +1711,12 @@ export interface components {
              * @default false
              */
             exportRemoteTraces?: boolean;
+            /**
+             * Attachuserid
+             * @description 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.
+             * @default false
+             */
+            attachUserId?: boolean;
             /** Contexts */
             contexts?: components["schemas"]["ChatContext"][];
             /**
@@ -2355,6 +2388,16 @@ export interface components {
              * @description The ID of the dataset version associated with the experiment
              */
             dataset_version_id: string;
+            /**
+             * Name
+             * @description The name of the experiment
+             */
+            name: string;
+            /**
+             * Description
+             * @description The description of the experiment
+             */
+            description: string | null;
             /**
              * Repetitions
              * @description Number of times the experiment is repeated
@@ -3175,6 +3218,7 @@ export interface components {
              * @default 1
              */
             repetitions?: number;
+            nextExperimentScaffold?: components["schemas"]["PlaygroundExperimentScaffoldContext"] | null;
             /** Instances */
             instances?: components["schemas"]["PlaygroundInstanceContext"][];
             /** Evaluators */
@@ -3219,6 +3263,37 @@ export interface components {
             /** Isapplied */
             isApplied: boolean;
         };
+        /**
+         * PlaygroundExperimentScaffoldContext
+         * @description 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.
+         */
+        PlaygroundExperimentScaffoldContext: {
+            /** Name */
+            name?: string | null;
+            /** Description */
+            description?: string | null;
+            /**
+             * Hasmetadata
+             * @default false
+             */
+            hasMetadata?: boolean;
+        };
         /**
          * PlaygroundInstanceContext
          * @description One mounted playground instance and its current model selection.
@@ -3460,6 +3535,19 @@ export interface components {
             /** Messages */
             messages: components["schemas"]["PromptMessage"][];
         };
+        /**
+         * PromptContext
+         * @description Prompt the user is currently viewing.
+         */
+        PromptContext: {
+            /**
+             * @description discriminator enum property added by openapi-typescript
+             * @enum {string}
+             */
+            type: "prompt";
+            /** Promptnodeid */
+            promptNodeId: string;
+        };
         /** PromptData */
         PromptData: {
             name: components["schemas"]["Identifier"];
@@ -3960,6 +4048,21 @@ export interface components {
             /** Id */
             id: string;
         };
+        /**
+         * PromptVersionContext
+         * @description Prompt version the user is currently viewing.
+         */
+        PromptVersionContext: {
+            /**
+             * @description discriminator enum property added by openapi-typescript
+             * @enum {string}
+             */
+            type: "prompt_version";
+            /** Promptnodeid */
+            promptNodeId: string;
+            /** Promptversionnodeid */
+            promptVersionNodeId: string;
+        };
         /** PromptVersionData */
         PromptVersionData: {
             /** Description */
@@ -4161,6 +4264,21 @@ export interface components {
             /** Next Cursor */
             next_cursor: string | null;
         };
+        /**
+         * SessionContext
+         * @description Session the user is currently viewing.
+         */
+        SessionContext: {
+            /**
+             * @description discriminator enum property added by openapi-typescript
+             * @enum {string}
+             */
+            type: "session";
+            /** Projectnodeid */
+            projectNodeId: string;
+            /** Sessionnodeid */
+            sessionNodeId: string;
+        };
         /** SessionData */
         SessionData: {
             /** Id */
@@ -4587,6 +4705,19 @@ export interface components {
              */
             type?: "step-start";
         };
+        /**
+         * SubagentsContext
+         * @description User's per-turn request to expose the subagent-spawning tool.
+         */
+        SubagentsContext: {
+            /**
+             * @description discriminator enum property added by openapi-typescript
+             * @enum {string}
+             */
+            type: "subagents";
+            /** Enabled */
+            enabled: boolean;
+        };
         /** TextContentPart */
         TextContentPart: {
             /**
@@ -5106,6 +5237,33 @@ export interface components {
             /** Data */
             data: components["schemas"]["CategoricalAnnotationConfig"] | components["schemas"]["ContinuousAnnotationConfig"] | components["schemas"]["FreeformAnnotationConfig"];
         };
+        /**
+         * UpdateExperimentRequestBody
+         * @description Fields to update on an experiment. Omit a field to leave it unchanged.
+         */
+        UpdateExperimentRequestBody: {
+            /**
+             * Name
+             * @description New name for the experiment (null is rejected; name is required)
+             */
+            name?: string | null;
+            /**
+             * Description
+             * @description New description for the experiment (null clears the description)
+             */
+            description?: string | null;
+            /**
+             * Metadata
+             * @description New metadata object for the experiment (replaces the existing metadata as a whole; null is rejected)
+             */
+            metadata?: {
+                [key: string]: unknown;
+            } | null;
+        };
+        /** UpdateExperimentResponseBody */
+        UpdateExperimentResponseBody: {
+            data: components["schemas"]["Experiment"];
+        };
         /** UpdateProjectRequestBody */
         UpdateProjectRequestBody: {
             /** Description */
@@ -5256,6 +5414,12 @@ export interface components {
              * @default false
              */
             exportRemoteTraces?: boolean;
+            /**
+             * Attachuserid
+             * @description 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.
+             * @default false
+             */
+            attachUserId?: boolean;
             /** Messages */
             messages: components["schemas"]["UIMessage"][];
             /** Model */
@@ -6578,6 +6742,59 @@ export interface operations {
             };
         };
     };
+    updateExperiment: {
+        parameters: {
+            query?: never;
+            header?: never;
+            path: {
+                experiment_id: string;
+            };
+            cookie?: never;
+        };
+        requestBody: {
+            content: {
+                "application/json": components["schemas"]["UpdateExperimentRequestBody"];
+            };
+        };
+        responses: {
+            /** @description Experiment updated successfully */
+            200: {
+                headers: {
+                    [name: string]: unknown;
+                };
+                content: {
+                    "application/json": components["schemas"]["UpdateExperimentResponseBody"];
+                };
+            };
+            /** @description Forbidden */
+            403: {
+                headers: {
+                    [name: string]: unknown;
+                };
+                content: {
+                    "text/plain": string;
+                };
+            };
+            /** @description Experiment not found */
+            404: {
+                headers: {
+                    [name: string]: unknown;
+                };
+                content: {
+                    "text/plain": string;
+                };
+            };
+            /** @description Invalid experiment ID or request body */
+            422: {
+                headers: {
+                    [name: string]: unknown;
+                };
+                content: {
+                    "text/plain": string;
+                };
+            };
+        };
+    };
     getIncompleteExperimentRuns: {
         parameters: {
             query?: {
@@ -9088,6 +9305,41 @@ export interface operations {
             };
         };
     };
+    run_server_agent_agents_server_sessions__session_id__chat_post: {
+        parameters: {
+            query?: never;
+            header?: never;
+            path: {
+                session_id: string;
+            };
+            cookie?: never;
+        };
+        requestBody: {
+            content: {
+                "application/json": components["schemas"]["ChatRequest"];
+            };
+        };
+        responses: {
+            /** @description Successful Response */
+            200: {
+                headers: {
+                    [name: string]: unknown;
+                };
+                content: {
+                    "application/json": unknown;
+                };
+            };
+            /** @description Validation Error */
+            422: {
+                headers: {
+                    [name: string]: unknown;
+                };
+                content: {
+                    "application/json": components["schemas"]["HTTPValidationError"];
+                };
+            };
+        };
+    };
     chat_agents__agent_id__sessions__session_id__chat_post: {
         parameters: {
             query?: never;

package/src/jest/index.ts

@@ -0,0 +1,124 @@
+import { createTestApi } from "../testing/define-api";
+import type { RunnerHooks } from "../testing/runner";
+
+export type {
+  PhoenixDescribe,
+  PhoenixTest,
+  PhoenixTestApi,
+  PhoenixTestEach,
+} from "../testing/define-api";
+
+export type {
+  AcceptanceCriterion,
+  AcceptanceMetric,
+  AcceptanceResult,
+  Annotation,
+  AnnotatorKind,
+  EvaluationParams,
+  EvaluationResult,
+  Evaluator,
+  EvaluatorResult,
+  KVMap,
+  ReferenceOutput,
+  SuiteConfig,
+  TestArgs,
+  TestConfig,
+  TestEachRow,
+  TestFn,
+  TestParams,
+  TestParamsBase,
+} from "../testing/types";
+
+export {
+  evaluate,
+  logAnnotation,
+  logOutput,
+  traceEvaluator,
+} from "../testing/helpers";
+
+/**
+ * Resolve jest globals lazily. Jest injects `describe`, `test`, `it`,
+ * `beforeAll`, `afterAll` onto the global object during test runs; we read
+ * them off `globalThis` so importing this module outside of a jest run
+ * doesn't error.
+ */
+function getJestGlobals(): {
+  describe: JestDescribe;
+  test: JestTest;
+  beforeAll: (fn: () => unknown | Promise<unknown>) => void;
+  afterAll: (fn: () => unknown | Promise<unknown>) => void;
+} {
+  const g = globalThis as unknown as {
+    describe?: JestDescribe;
+    test?: JestTest;
+    it?: JestTest;
+    beforeAll?: (fn: () => unknown | Promise<unknown>) => void;
+    afterAll?: (fn: () => unknown | Promise<unknown>) => void;
+  };
+  if (!g.describe || !g.beforeAll || !g.afterAll) {
+    throw new Error(
+      "@arizeai/phoenix-client/jest could not find Jest globals. Make sure this module " +
+        "is imported from inside a Jest test file."
+    );
+  }
+  const test = g.test ?? g.it;
+  if (!test) {
+    throw new Error(
+      "@arizeai/phoenix-client/jest could not find Jest's test/it."
+    );
+  }
+  return {
+    describe: g.describe,
+    test,
+    beforeAll: g.beforeAll,
+    afterAll: g.afterAll,
+  };
+}
+
+interface JestDescribe {
+  (name: string, fn: () => void): void;
+  only: (name: string, fn: () => void) => void;
+  skip: (name: string, fn: () => void) => void;
+}
+
+interface JestTest {
+  (
+    name: string,
+    fn: (...args: unknown[]) => unknown | Promise<unknown>,
+    timeout?: number
+  ): void;
+  only: (
+    name: string,
+    fn: (...args: unknown[]) => unknown | Promise<unknown>,
+    timeout?: number
+  ) => void;
+  skip: (
+    name: string,
+    fn: (...args: unknown[]) => unknown | Promise<unknown>,
+    timeout?: number
+  ) => void;
+}
+
+/**
+ * Build the runner hooks from the (lazily-resolved) jest globals. Memoized so
+ * repeated declarations don't re-read `globalThis` on every call; jest injects
+ * its globals once per worker before any test file is evaluated.
+ */
+let cachedHooks: RunnerHooks | undefined;
+function getHooks(): RunnerHooks {
+  if (cachedHooks) return cachedHooks;
+  const g = getJestGlobals();
+  cachedHooks = {
+    describe: (name, fn) => g.describe(name, fn),
+    describeOnly: (name, fn) => g.describe.only(name, fn),
+    describeSkip: (name, fn) => g.describe.skip(name, fn),
+    test: (name, fn, timeout) => g.test(name, fn, timeout),
+    testOnly: (name, fn, timeout) => g.test.only(name, fn, timeout),
+    testSkip: (name, fn, timeout) => g.test.skip(name, fn, timeout),
+    beforeAll: (fn) => g.beforeAll(fn),
+    afterAll: (fn) => g.afterAll(fn),
+  };
+  return cachedHooks;
+}
+
+export const { describe, test, it } = createTestApi(getHooks);

package/src/jest/reporter.ts

@@ -0,0 +1,22 @@
+import { SuiteSummaryReportRun } from "../testing/report-run";
+
+/**
+ * Jest reporter for `@arizeai/phoenix-client/jest`.
+ *
+ * Like the Vitest reporter, this does not replace Jest's default output. It
+ * appends a Phoenix summary block at the end of the run.
+ */
+export default class PhoenixJestReporter {
+  private readonly report = new SuiteSummaryReportRun();
+
+  // jest passes globalConfig + reporterOptions; we ignore both
+  constructor(_globalConfig?: unknown, _reporterOptions?: unknown) {}
+
+  onRunStart(): void {
+    this.report.begin();
+  }
+
+  onRunComplete(): void {
+    this.report.finish();
+  }
+}

package/src/prompts/sdks/toAI.ts

@@ -40,7 +40,8 @@ The tool choice strategy. Default: 'auto'.
   toolChoice?: ToolChoice<ToolSet>;
 };
 
-export type ToAIParams<V extends Variables> = toSDKParamsBase<V>;
+export type ToAIParams<PromptVariables extends Variables> =
+  toSDKParamsBase<PromptVariables>;
 
 /**
  * Converts a Phoenix prompt to Vercel AI sdk params.
@@ -48,10 +49,10 @@ export type ToAIParams<V extends Variables> = toSDKParamsBase<V>;
  * - note: To use response format, you must pass `prompt.response_format.json_schema.schema` to generateObject or streamObject
  *   via `jsonSchema()`, through the `schema` argument.
  */
-export const toAI = <V extends Variables>({
+export const toAI = <PromptVariables extends Variables>({
   prompt,
   variables,
-}: ToAIParams<V>): PartialAIParams | null => {
+}: ToAIParams<PromptVariables>): PartialAIParams | null => {
   // eslint-disable-next-line no-console
   console.warn(
     "Prompt invocation parameters not currently supported in AI SDK, falling back to default invocation parameters"

package/src/prompts/sdks/toAnthropic.ts

@@ -14,15 +14,16 @@ import type { toSDKParamsBase, Variables } from "./types";
 // We must re-export these types so that they are included in the phoenix-client distribution
 export type { MessageCreateParams };
 
-export type ToAnthropicParams<V extends Variables> = toSDKParamsBase<V>;
+export type ToAnthropicParams<PromptVariables extends Variables> =
+  toSDKParamsBase<PromptVariables>;
 
 /**
  * Convert a Phoenix prompt to Anthropic client sdk's message create parameters
  */
-export const toAnthropic = <V extends Variables = Variables>({
+export const toAnthropic = <PromptVariables extends Variables = Variables>({
   prompt,
   variables,
-}: ToAnthropicParams<V>): MessageCreateParams | null => {
+}: ToAnthropicParams<PromptVariables>): MessageCreateParams | null => {
   try {
     let invocationParameters: { max_tokens: number } | undefined;
     if (prompt.invocation_parameters.type === "anthropic") {

package/src/prompts/sdks/toOpenAI.ts

@@ -23,17 +23,18 @@ export type {
   ResponseFormatJSONSchema,
 };
 
-export type ToOpenAIParams<V extends Variables> = toSDKParamsBase<V>;
+export type ToOpenAIParams<PromptVariables extends Variables> =
+  toSDKParamsBase<PromptVariables>;
 
 /**
  * Convert a Phoenix prompt to OpenAI client sdk's chat completion parameters
  *
  * @returns The converted chat completion parameters
  */
-export const toOpenAI = <V extends Variables = Variables>({
+export const toOpenAI = <PromptVariables extends Variables = Variables>({
   prompt,
   variables,
-}: ToOpenAIParams<V>): ChatCompletionCreateParams | null => {
+}: ToOpenAIParams<PromptVariables>): ChatCompletionCreateParams | null => {
   try {
     let invocationParameters: Partial<ChatCompletionCreateParams>;
     switch (prompt.invocation_parameters.type) {

package/src/prompts/sdks/toSDK.ts

@@ -15,10 +15,10 @@ import type { SupportedSDK, toSDKParamsBase, Variables } from "./types";
  * toOpenAI(params)
  * ```
  */
-export type SDKParams<T extends SupportedSDK> = Parameters<
-  (typeof PROVIDER_TO_SDK)[T]
+export type SDKParams<SDK extends SupportedSDK> = Parameters<
+  (typeof PROVIDER_TO_SDK)[SDK]
 >[number] extends toSDKParamsBase
-  ? Parameters<(typeof PROVIDER_TO_SDK)[T]>[number]
+  ? Parameters<(typeof PROVIDER_TO_SDK)[SDK]>[number]
   : never;
 
 /**
@@ -33,7 +33,7 @@ export const PROVIDER_TO_SDK = {
 /**
  * Get the conversion function for a specific SDK name
  */
-const getTargetSDK = <T extends SupportedSDK>(sdk: T) => {
+const getTargetSDK = <SDK extends SupportedSDK>(sdk: SDK) => {
   switch (sdk) {
     case "openai":
       return PROVIDER_TO_SDK.openai;
@@ -50,19 +50,19 @@ const getTargetSDK = <T extends SupportedSDK>(sdk: T) => {
  * Parameters specific to the toSDK function
  */
 export type ToSDKParams<
-  T extends SupportedSDK,
-  V extends Variables = Variables,
+  SDK extends SupportedSDK,
+  PromptVariables extends Variables = Variables,
 > = {
   /**
    * String representing the SDK to convert to
    */
-  sdk: T;
+  sdk: SDK;
   /**
    * Optional variables to format the prompt with
    * Keys are the variable names, values are the variable values
    * The variable format is determined via prompt.template_format
    */
-  variables?: V;
+  variables?: PromptVariables;
 };
 
 /**
@@ -86,11 +86,16 @@ export type ToSDKParams<
  * @param params - The parameters to convert a prompt to an SDK's parameters
  * @returns The SDK's parameters
  */
-export const toSDK = <T extends SupportedSDK, V extends Variables = Variables>({
+export const toSDK = <
+  SDK extends SupportedSDK,
+  PromptVariables extends Variables = Variables,
+>({
   sdk: _sdk,
   ...rest
-}: ToSDKParams<T, V> & SDKParams<T>) => {
+}: ToSDKParams<SDK, PromptVariables> & SDKParams<SDK>) => {
   const sdk = getTargetSDK(_sdk);
   invariant(sdk, `No SDK found for provider ${_sdk}`);
-  return sdk<V>(rest) as ReturnType<(typeof PROVIDER_TO_SDK)[T]>;
+  return sdk<PromptVariables>(rest) as ReturnType<
+    (typeof PROVIDER_TO_SDK)[SDK]
+  >;
 };

package/src/prompts/sdks/types.ts

@@ -11,7 +11,7 @@ export type Variables = Record<string, string | { toString: () => string }>;
 /**
  * Base parameters for an SDK conversion function
  */
-export type toSDKParamsBase<V extends Variables = Variables> = {
+export type toSDKParamsBase<PromptVariables extends Variables = Variables> = {
   /**
    * The Phoenix prompt to convert
    */
@@ -19,7 +19,7 @@ export type toSDKParamsBase<V extends Variables = Variables> = {
   /**
    * The variables to use in the prompt
    */
-  variables?: V;
+  variables?: PromptVariables;
 };
 
 /**