@meetkai/mka1 0.54.22 → 0.54.24

package/src/models/components/importagentrunrequest.ts

@@ -0,0 +1,319 @@
+/*
+ * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
+ */
+
+import * as z from "zod/v3";
+import { remap as remap$ } from "../../lib/primitives.js";
+import { safeParse } from "../../lib/schemas.js";
+import { ClosedEnum } from "../../types/enums.js";
+import { Result as SafeParseResult } from "../../types/fp.js";
+import { SDKValidationError } from "../errors/sdkvalidationerror.js";
+
+/**
+ * `failed` means no score was produced at all. An agent timeout is NOT that — the harness still runs the verifier and scores the trial, so it stays `completed` and carries its exception in `error`.
+ */
+export const ImportAgentRunRequestStatus = {
+  Completed: "completed",
+  Failed: "failed",
+} as const;
+/**
+ * `failed` means no score was produced at all. An agent timeout is NOT that — the harness still runs the verifier and scores the trial, so it stays `completed` and carries its exception in `error`.
+ */
+export type ImportAgentRunRequestStatus = ClosedEnum<
+  typeof ImportAgentRunRequestStatus
+>;
+
+export type ImportAgentRunRequestTrials = {
+  /**
+   * Must already exist in the target suite version's manifest; unknown ids are rejected rather than invented.
+   */
+  taskId: string;
+  /**
+   * Defaults to the trial's position in `trials`. Unique per (task, model), not per run.
+   */
+  sampleIndex?: number | undefined;
+  /**
+   * `failed` means no score was produced at all. An agent timeout is NOT that — the harness still runs the verifier and scores the trial, so it stays `completed` and carries its exception in `error`.
+   */
+  status?: ImportAgentRunRequestStatus | undefined;
+  /**
+   * The task instruction.
+   */
+  prompt?: string | undefined;
+  target?: string | null | undefined;
+  /**
+   * What decided the score — the verifier's output. The agent's session goes to the trajectory route.
+   */
+  outputText?: string | null | undefined;
+  /**
+   * Metric values for this trial, e.g. `{ "reward": 1, "resolved": 1 }`. The board averages `reward` by default.
+   */
+  scores?: { [k: string]: number } | undefined;
+  /**
+   * The exception this trial recorded, if any. Does not by itself make the trial a failure.
+   */
+  error?: { [k: string]: any } | null | undefined;
+  costUsd?: number | null | undefined;
+  inputTokens?: number | null | undefined;
+  outputTokens?: number | null | undefined;
+  /**
+   * The share of `input_tokens` served from the provider's prompt cache.
+   */
+  cachedTokens?: number | null | undefined;
+  phaseTimings?: { [k: string]: any } | null | undefined;
+};
+
+export type ImportAgentRunRequest = {
+  /**
+   * Existing suite to import into. Its manifest defines the task set, so create the suite first.
+   */
+  suiteId: string;
+  /**
+   * Defaults to the suite's active version.
+   */
+  suiteVersion?: number | undefined;
+  /**
+   * The harness agent, e.g. `omp`.
+   */
+  agentName: string;
+  agentVersion?: string | null | undefined;
+  /**
+   * Reasoning-effort setting, e.g. `medium`. Part of the leaderboard's row identity, so the same agent and model at two efforts stay two rows.
+   */
+  agentEffort?: string | null | undefined;
+  model: string;
+  /**
+   * Total spend for the job. Falls back to the sum of the trials' own costs when omitted.
+   */
+  costUsd?: number | null | undefined;
+  /**
+   * When the job ran, unix seconds. Defaults to now — set it when backfilling, since the board orders by this.
+   */
+  createdAt?: number | undefined;
+  metadata?: { [k: string]: string } | undefined;
+  /**
+   * Reusing a key returns the run created the first time instead of importing a second copy. Without one, a retried upload silently doubles the trials behind a board row.
+   */
+  idempotencyKey?: string | undefined;
+  /**
+   * May be empty when opening a run to stream trials into. A one-shot import sends them all here.
+   */
+  trials?: Array<ImportAgentRunRequestTrials> | undefined;
+  /**
+   * How many trials the job will produce. Sets the run's total up front so the console shows real progress (12/89) while it streams, instead of 12/12 that grows.
+   */
+  expectedTrials?: number | undefined;
+  /**
+   * False opens the run `in_progress` for `POST /evals/runs/{run_id}/trials` to fill. Leaderboards only count completed runs, so a streaming run stays off the board until it is finalized.
+   */
+  finalize?: boolean | undefined;
+};
+
+/** @internal */
+export const ImportAgentRunRequestStatus$inboundSchema: z.ZodNativeEnum<
+  typeof ImportAgentRunRequestStatus
+> = z.nativeEnum(ImportAgentRunRequestStatus);
+/** @internal */
+export const ImportAgentRunRequestStatus$outboundSchema: z.ZodNativeEnum<
+  typeof ImportAgentRunRequestStatus
+> = ImportAgentRunRequestStatus$inboundSchema;
+
+/** @internal */
+export const ImportAgentRunRequestTrials$inboundSchema: z.ZodType<
+  ImportAgentRunRequestTrials,
+  z.ZodTypeDef,
+  unknown
+> = z.object({
+  task_id: z.string(),
+  sample_index: z.number().int().optional(),
+  status: ImportAgentRunRequestStatus$inboundSchema.default("completed"),
+  prompt: z.string().default(""),
+  target: z.nullable(z.string()).optional(),
+  output_text: z.nullable(z.string()).optional(),
+  scores: z.record(z.number()).optional(),
+  error: z.nullable(z.record(z.any())).optional(),
+  cost_usd: z.nullable(z.number()).optional(),
+  input_tokens: z.nullable(z.number().int()).optional(),
+  output_tokens: z.nullable(z.number().int()).optional(),
+  cached_tokens: z.nullable(z.number().int()).optional(),
+  phase_timings: z.nullable(z.record(z.any())).optional(),
+}).transform((v) => {
+  return remap$(v, {
+    "task_id": "taskId",
+    "sample_index": "sampleIndex",
+    "output_text": "outputText",
+    "cost_usd": "costUsd",
+    "input_tokens": "inputTokens",
+    "output_tokens": "outputTokens",
+    "cached_tokens": "cachedTokens",
+    "phase_timings": "phaseTimings",
+  });
+});
+/** @internal */
+export type ImportAgentRunRequestTrials$Outbound = {
+  task_id: string;
+  sample_index?: number | undefined;
+  status: string;
+  prompt: string;
+  target?: string | null | undefined;
+  output_text?: string | null | undefined;
+  scores?: { [k: string]: number } | undefined;
+  error?: { [k: string]: any } | null | undefined;
+  cost_usd?: number | null | undefined;
+  input_tokens?: number | null | undefined;
+  output_tokens?: number | null | undefined;
+  cached_tokens?: number | null | undefined;
+  phase_timings?: { [k: string]: any } | null | undefined;
+};
+
+/** @internal */
+export const ImportAgentRunRequestTrials$outboundSchema: z.ZodType<
+  ImportAgentRunRequestTrials$Outbound,
+  z.ZodTypeDef,
+  ImportAgentRunRequestTrials
+> = z.object({
+  taskId: z.string(),
+  sampleIndex: z.number().int().optional(),
+  status: ImportAgentRunRequestStatus$outboundSchema.default("completed"),
+  prompt: z.string().default(""),
+  target: z.nullable(z.string()).optional(),
+  outputText: z.nullable(z.string()).optional(),
+  scores: z.record(z.number()).optional(),
+  error: z.nullable(z.record(z.any())).optional(),
+  costUsd: z.nullable(z.number()).optional(),
+  inputTokens: z.nullable(z.number().int()).optional(),
+  outputTokens: z.nullable(z.number().int()).optional(),
+  cachedTokens: z.nullable(z.number().int()).optional(),
+  phaseTimings: z.nullable(z.record(z.any())).optional(),
+}).transform((v) => {
+  return remap$(v, {
+    taskId: "task_id",
+    sampleIndex: "sample_index",
+    outputText: "output_text",
+    costUsd: "cost_usd",
+    inputTokens: "input_tokens",
+    outputTokens: "output_tokens",
+    cachedTokens: "cached_tokens",
+    phaseTimings: "phase_timings",
+  });
+});
+
+export function importAgentRunRequestTrialsToJSON(
+  importAgentRunRequestTrials: ImportAgentRunRequestTrials,
+): string {
+  return JSON.stringify(
+    ImportAgentRunRequestTrials$outboundSchema.parse(
+      importAgentRunRequestTrials,
+    ),
+  );
+}
+export function importAgentRunRequestTrialsFromJSON(
+  jsonString: string,
+): SafeParseResult<ImportAgentRunRequestTrials, SDKValidationError> {
+  return safeParse(
+    jsonString,
+    (x) => ImportAgentRunRequestTrials$inboundSchema.parse(JSON.parse(x)),
+    `Failed to parse 'ImportAgentRunRequestTrials' from JSON`,
+  );
+}
+
+/** @internal */
+export const ImportAgentRunRequest$inboundSchema: z.ZodType<
+  ImportAgentRunRequest,
+  z.ZodTypeDef,
+  unknown
+> = z.object({
+  suite_id: z.string(),
+  suite_version: z.number().int().optional(),
+  agent_name: z.string(),
+  agent_version: z.nullable(z.string()).optional(),
+  agent_effort: z.nullable(z.string()).optional(),
+  model: z.string(),
+  cost_usd: z.nullable(z.number()).optional(),
+  created_at: z.number().int().optional(),
+  metadata: z.record(z.string()).optional(),
+  idempotency_key: z.string().optional(),
+  trials: z.array(z.lazy(() => ImportAgentRunRequestTrials$inboundSchema))
+    .optional(),
+  expected_trials: z.number().int().optional(),
+  finalize: z.boolean().default(true),
+}).transform((v) => {
+  return remap$(v, {
+    "suite_id": "suiteId",
+    "suite_version": "suiteVersion",
+    "agent_name": "agentName",
+    "agent_version": "agentVersion",
+    "agent_effort": "agentEffort",
+    "cost_usd": "costUsd",
+    "created_at": "createdAt",
+    "idempotency_key": "idempotencyKey",
+    "expected_trials": "expectedTrials",
+  });
+});
+/** @internal */
+export type ImportAgentRunRequest$Outbound = {
+  suite_id: string;
+  suite_version?: number | undefined;
+  agent_name: string;
+  agent_version?: string | null | undefined;
+  agent_effort?: string | null | undefined;
+  model: string;
+  cost_usd?: number | null | undefined;
+  created_at?: number | undefined;
+  metadata?: { [k: string]: string } | undefined;
+  idempotency_key?: string | undefined;
+  trials?: Array<ImportAgentRunRequestTrials$Outbound> | undefined;
+  expected_trials?: number | undefined;
+  finalize: boolean;
+};
+
+/** @internal */
+export const ImportAgentRunRequest$outboundSchema: z.ZodType<
+  ImportAgentRunRequest$Outbound,
+  z.ZodTypeDef,
+  ImportAgentRunRequest
+> = z.object({
+  suiteId: z.string(),
+  suiteVersion: z.number().int().optional(),
+  agentName: z.string(),
+  agentVersion: z.nullable(z.string()).optional(),
+  agentEffort: z.nullable(z.string()).optional(),
+  model: z.string(),
+  costUsd: z.nullable(z.number()).optional(),
+  createdAt: z.number().int().optional(),
+  metadata: z.record(z.string()).optional(),
+  idempotencyKey: z.string().optional(),
+  trials: z.array(z.lazy(() => ImportAgentRunRequestTrials$outboundSchema))
+    .optional(),
+  expectedTrials: z.number().int().optional(),
+  finalize: z.boolean().default(true),
+}).transform((v) => {
+  return remap$(v, {
+    suiteId: "suite_id",
+    suiteVersion: "suite_version",
+    agentName: "agent_name",
+    agentVersion: "agent_version",
+    agentEffort: "agent_effort",
+    costUsd: "cost_usd",
+    createdAt: "created_at",
+    idempotencyKey: "idempotency_key",
+    expectedTrials: "expected_trials",
+  });
+});
+
+export function importAgentRunRequestToJSON(
+  importAgentRunRequest: ImportAgentRunRequest,
+): string {
+  return JSON.stringify(
+    ImportAgentRunRequest$outboundSchema.parse(importAgentRunRequest),
+  );
+}
+export function importAgentRunRequestFromJSON(
+  jsonString: string,
+): SafeParseResult<ImportAgentRunRequest, SDKValidationError> {
+  return safeParse(
+    jsonString,
+    (x) => ImportAgentRunRequest$inboundSchema.parse(JSON.parse(x)),
+    `Failed to parse 'ImportAgentRunRequest' from JSON`,
+  );
+}

package/src/models/components/index.ts

@@ -27,6 +27,7 @@ export * from "./agenttextconfiguration.js";
 export * from "./agentversion.js";
 export * from "./agentversionlist.js";
 export * from "./allowedtools.js";
+export * from "./appendagenttrialsrequest.js";
 export * from "./authorizeintent.js";
 export * from "./autochunkingstrategy.js";
 export * from "./autoendpoint.js";
@@ -233,6 +234,7 @@ export * from "./evaldatasetformat.js";
 export * from "./evalfewshotconfig.js";
 export * from "./evalgenerationconfig.js";
 export * from "./evalhistoricalimportobject.js";
+export * from "./evalleaderboardrow.js";
 export * from "./evalleaderboardrungroup.js";
 export * from "./evalmetric.js";
 export * from "./evaloutputextraction.js";
@@ -242,15 +244,18 @@ export * from "./evalpythongraderexecution.js";
 export * from "./evalpythongradermodelaccess.js";
 export * from "./evalpythonpreprocessor.js";
 export * from "./evalrundeletedobject.js";
+export * from "./evalrunoverview.js";
 export * from "./evalrunstatus.js";
 export * from "./evalsampleaudioobject.js";
 export * from "./evalsamplelist.js";
 export * from "./evalsampleobject.js";
 export * from "./evalsamplestatus.js";
+export * from "./evalsampletrajectoryuploadedobject.js";
 export * from "./evalscheduledeletedobject.js";
 export * from "./evalscheduleoverlappolicy.js";
 export * from "./evalschedulerunconfig.js";
 export * from "./evalsuitedeletedobject.js";
+export * from "./evalsuiteleaderboard.js";
 export * from "./evalsuitelist.js";
 export * from "./evalsuitemanifest.js";
 export * from "./evalsuiteobject.js";
@@ -309,6 +314,7 @@ export * from "./imagegenerationcallwithid.js";
 export * from "./imagegenerationrequest.js";
 export * from "./imagegenerationresponse.js";
 export * from "./imagegenerationtooldefinition.js";
+export * from "./importagentrunrequest.js";
 export * from "./importhistoricalevalresultsrequest.js";
 export * from "./incompletedetails.js";
 export * from "./inputaudio.js";
@@ -494,6 +500,7 @@ export * from "./responsereasoningtextdeltaevent.js";
 export * from "./responsereasoningtextdoneevent.js";
 export * from "./responserefusaldeltaevent.js";
 export * from "./responserefusaldoneevent.js";
+export * from "./responsescreaterequest.js";
 export * from "./responsesgroupby.js";
 export * from "./responseshellcallcompletedevent.js";
 export * from "./responseshellcallexecutingevent.js";
@@ -606,6 +613,7 @@ export * from "./updatememorystorerequest.js";
 export * from "./updatepromptresponse.js";
 export * from "./updateresponserequest.js";
 export * from "./updatevectorstorefileattributesrequest.js";
+export * from "./uploadevalsampletrajectoryrequest.js";
 export * from "./urlcitation.js";
 export * from "./usage.js";
 export * from "./usagecostsresponse.js";