@meetkai/mka1 0.54.23 → 0.54.25

package/dist/esm/funcs/llmEvalsGetSuiteLeaderboard.js

@@ -0,0 +1,106 @@
+/*
+ * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
+ */
+import { encodeFormQuery, encodeSimple } from "../lib/encodings.js";
+import { matchStatusCode } from "../lib/http.js";
+import * as M from "../lib/matchers.js";
+import { compactMap } from "../lib/primitives.js";
+import { safeParse } from "../lib/schemas.js";
+import { extractSecurity, resolveGlobalSecurity } from "../lib/security.js";
+import { pathToFunc } from "../lib/url.js";
+import * as components from "../models/components/index.js";
+import * as operations from "../models/operations/index.js";
+import { APIPromise } from "../types/async.js";
+/**
+ * Get the coding-agent leaderboard for an eval suite
+ *
+ * @remarks
+ * Ranks the suite's completed coding-agent runs, one row per (agent, model, effort). Only runs that recorded an `agent_name` participate — a QA or transcription run of the same suite has no agent to attribute a score to and is not a board row. `accuracy` is the mean of `metric` (default `reward`) over the row's scored trials, and `accuracy_ci` is the 95% Wald interval around it; both are derived per request rather than stored, so rerunning a failed trial moves them instead of leaving a stale copy behind. `agent_version` is reported but is not part of the row identity, so an agent upgrade does not split a row in two. Rows are ordered by accuracy descending, nulls last. Pass `suite_version` when the suite has more than one, otherwise the board mixes task sets.
+ *
+ * If set, this operation will use {@link Security.bearerAuth} from the global security.
+ */
+export function llmEvalsGetSuiteLeaderboard(client, request, options) {
+    return new APIPromise($do(client, request, options));
+}
+async function $do(client, request, options) {
+    const parsed = safeParse(request, (value) => operations.GetEvalSuiteLeaderboardRequest$outboundSchema.parse(value), "Input validation failed");
+    if (!parsed.ok) {
+        return [parsed, { status: "invalid" }];
+    }
+    const payload = parsed.value;
+    const body = null;
+    const pathParams = {
+        suite_id: encodeSimple("suite_id", payload.suite_id, {
+            explode: false,
+            charEncoding: "percent",
+        }),
+    };
+    const path = pathToFunc("/api/v1/llm/evals/suites/{suite_id}/leaderboard")(pathParams);
+    const query = encodeFormQuery({
+        "agent": payload.agent,
+        "limit": payload.limit,
+        "metric": payload.metric,
+        "model": payload.model,
+        "suite_version": payload.suite_version,
+    });
+    const headers = new Headers(compactMap({
+        Accept: "application/json",
+        "X-On-Behalf-Of": encodeSimple("X-On-Behalf-Of", payload["X-On-Behalf-Of"], { explode: false, charEncoding: "none" }),
+    }));
+    const secConfig = await extractSecurity(client._options.bearerAuth);
+    const securityInput = secConfig == null ? {} : { bearerAuth: secConfig };
+    const requestSecurity = resolveGlobalSecurity(securityInput, [0]);
+    const context = {
+        options: client._options,
+        baseURL: options?.serverURL ?? client._baseURL ?? "",
+        operationID: "getEvalSuiteLeaderboard",
+        oAuth2Scopes: null,
+        resolvedSecurity: requestSecurity,
+        securitySource: client._options.bearerAuth,
+        retryConfig: options?.retries
+            || client._options.retryConfig
+            || {
+                strategy: "backoff",
+                backoff: {
+                    initialInterval: 500,
+                    maxInterval: 5000,
+                    exponent: 1.5,
+                    maxElapsedTime: 15000,
+                },
+                retryConnectionErrors: true,
+            }
+            || { strategy: "none" },
+        retryCodes: options?.retryCodes || ["429", "500", "502", "503", "504"],
+    };
+    const requestRes = client._createRequest(context, {
+        security: requestSecurity,
+        method: "GET",
+        baseURL: options?.serverURL,
+        path: path,
+        headers: headers,
+        query: query,
+        body: body,
+        userAgent: client._options.userAgent,
+        timeoutMs: options?.timeoutMs || client._options.timeoutMs || -1,
+    }, options);
+    if (!requestRes.ok) {
+        return [requestRes, { status: "invalid" }];
+    }
+    const req = requestRes.value;
+    const doResult = await client._do(req, {
+        context,
+        isErrorStatusCode: (statusCode) => matchStatusCode({ status: statusCode }, ["4XX", "5XX"]),
+        retryConfig: context.retryConfig,
+        retryCodes: context.retryCodes,
+    });
+    if (!doResult.ok) {
+        return [doResult, { status: "request-error", request: req }];
+    }
+    const response = doResult.value;
+    const [result] = await M.match(M.json(200, components.EvalSuiteLeaderboard$inboundSchema), M.fail("4XX"), M.fail("5XX"))(response, req);
+    if (!result.ok) {
+        return [result, { status: "complete", request: req, response }];
+    }
+    return [result, { status: "complete", request: req, response }];
+}
+//# sourceMappingURL=llmEvalsGetSuiteLeaderboard.js.map

package/dist/esm/funcs/llmEvalsGetSuiteLeaderboard.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"llmEvalsGetSuiteLeaderboard.js","sourceRoot":"","sources":["../../../src/funcs/llmEvalsGetSuiteLeaderboard.ts"],"names":[],"mappings":"AAAA;;GAEG;AAGH,OAAO,EAAE,eAAe,EAAE,YAAY,EAAE,MAAM,qBAAqB,CAAC;AACpE,OAAO,EAAE,eAAe,EAAE,MAAM,gBAAgB,CAAC;AACjD,OAAO,KAAK,CAAC,MAAM,oBAAoB,CAAC;AACxC,OAAO,EAAE,UAAU,EAAE,MAAM,sBAAsB,CAAC;AAClD,OAAO,EAAE,SAAS,EAAE,MAAM,mBAAmB,CAAC;AAE9C,OAAO,EAAE,eAAe,EAAE,qBAAqB,EAAE,MAAM,oBAAoB,CAAC;AAC5E,OAAO,EAAE,UAAU,EAAE,MAAM,eAAe,CAAC;AAC3C,OAAO,KAAK,UAAU,MAAM,+BAA+B,CAAC;AAW5D,OAAO,KAAK,UAAU,MAAM,+BAA+B,CAAC;AAC5D,OAAO,EAAW,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAGxD;;;;;;;GAOG;AACH,MAAM,UAAU,2BAA2B,CACzC,MAAe,EACf,OAAkD,EAClD,OAAwB;IAcxB,OAAO,IAAI,UAAU,CAAC,GAAG,CACvB,MAAM,EACN,OAAO,EACP,OAAO,CACR,CAAC,CAAC;AACL,CAAC;AAED,KAAK,UAAU,GAAG,CAChB,MAAe,EACf,OAAkD,EAClD,OAAwB;IAiBxB,MAAM,MAAM,GAAG,SAAS,CACtB,OAAO,EACP,CAAC,KAAK,EAAE,EAAE,CACR,UAAU,CAAC,6CAA6C,CAAC,KAAK,CAAC,KAAK,CAAC,EACvE,yBAAyB,CAC1B,CAAC;IACF,IAAI,CAAC,MAAM,CAAC,EAAE,EAAE,CAAC;QACf,OAAO,CAAC,MAAM,EAAE,EAAE,MAAM,EAAE,SAAS,EAAE,CAAC,CAAC;IACzC,CAAC;IACD,MAAM,OAAO,GAAG,MAAM,CAAC,KAAK,CAAC;IAC7B,MAAM,IAAI,GAAG,IAAI,CAAC;IAElB,MAAM,UAAU,GAAG;QACjB,QAAQ,EAAE,YAAY,CAAC,UAAU,EAAE,OAAO,CAAC,QAAQ,EAAE;YACnD,OAAO,EAAE,KAAK;YACd,YAAY,EAAE,SAAS;SACxB,CAAC;KACH,CAAC;IACF,MAAM,IAAI,GAAG,UAAU,CAAC,iDAAiD,CAAC,CACxE,UAAU,CACX,CAAC;IAEF,MAAM,KAAK,GAAG,eAAe,CAAC;QAC5B,OAAO,EAAE,OAAO,CAAC,KAAK;QACtB,OAAO,EAAE,OAAO,CAAC,KAAK;QACtB,QAAQ,EAAE,OAAO,CAAC,MAAM;QACxB,OAAO,EAAE,OAAO,CAAC,KAAK;QACtB,eAAe,EAAE,OAAO,CAAC,aAAa;KACvC,CAAC,CAAC;IAEH,MAAM,OAAO,GAAG,IAAI,OAAO,CAAC,UAAU,CAAC;QACrC,MAAM,EAAE,kBAAkB;QAC1B,gBAAgB,EAAE,YAAY,CAC5B,gBAAgB,EAChB,OAAO,CAAC,gBAAgB,CAAC,EACzB,EAAE,OAAO,EAAE,KAAK,EAAE,YAAY,EAAE,MAAM,EAAE,CACzC;KACF,CAAC,CAAC,CAAC;IAEJ,MAAM,SAAS,GAAG,MAAM,eAAe,CAAC,MAAM,CAAC,QAAQ,CAAC,UAAU,CAAC,CAAC;IACpE,MAAM,aAAa,GAAG,SAAS,IAAI,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,SAAS,EAAE,CAAC;IACzE,MAAM,eAAe,GAAG,qBAAqB,CAAC,aAAa,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;IAElE,MAAM,OAAO,GAAG;QACd,OAAO,EAAE,MAAM,CAAC,QAAQ;QACxB,OAAO,EAAE,OAAO,EAAE,SAAS,IAAI,MAAM,CAAC,QAAQ,IAAI,EAAE;QACpD,WAAW,EAAE,yBAAyB;QACtC,YAAY,EAAE,IAAI;QAElB,gBAAgB,EAAE,eAAe;QAEjC,cAAc,EAAE,MAAM,CAAC,QAAQ,CAAC,UAAU;QAC1C,WAAW,EAAE,OAAO,EAAE,OAAO;eACxB,MAAM,CAAC,QAAQ,CAAC,WAAW;eAC3B;gBACD,QAAQ,EAAE,SAAS;gBACnB,OAAO,EAAE;oBACP,eAAe,EAAE,GAAG;oBACpB,WAAW,EAAE,IAAI;oBACjB,QAAQ,EAAE,GAAG;oBACb,cAAc,EAAE,KAAK;iBACtB;gBACD,qBAAqB,EAAE,IAAI;aAC5B;eACE,EAAE,QAAQ,EAAE,MAAM,EAAE;QACzB,UAAU,EAAE,OAAO,EAAE,UAAU,IAAI,CAAC,KAAK,EAAE,KAAK,EAAE,KAAK,EAAE,KAAK,EAAE,KAAK,CAAC;KACvE,CAAC;IAEF,MAAM,UAAU,GAAG,MAAM,CAAC,cAAc,CAAC,OAAO,EAAE;QAChD,QAAQ,EAAE,eAAe;QACzB,MAAM,EAAE,KAAK;QACb,OAAO,EAAE,OAAO,EAAE,SAAS;QAC3B,IAAI,EAAE,IAAI;QACV,OAAO,EAAE,OAAO;QAChB,KAAK,EAAE,KAAK;QACZ,IAAI,EAAE,IAAI;QACV,SAAS,EAAE,MAAM,CAAC,QAAQ,CAAC,SAAS;QACpC,SAAS,EAAE,OAAO,EAAE,SAAS,IAAI,MAAM,CAAC,QAAQ,CAAC,SAAS,IAAI,CAAC,CAAC;KACjE,EAAE,OAAO,CAAC,CAAC;IACZ,IAAI,CAAC,UAAU,CAAC,EAAE,EAAE,CAAC;QACnB,OAAO,CAAC,UAAU,EAAE,EAAE,MAAM,EAAE,SAAS,EAAE,CAAC,CAAC;IAC7C,CAAC;IACD,MAAM,GAAG,GAAG,UAAU,CAAC,KAAK,CAAC;IAE7B,MAAM,QAAQ,GAAG,MAAM,MAAM,CAAC,GAAG,CAAC,GAAG,EAAE;QACrC,OAAO;QACP,iBAAiB,EAAE,CAAC,UAAkB,EAAE,EAAE,CACxC,eAAe,CAAC,EAAE,MAAM,EAAE,UAAU,EAAc,EAAE,CAAC,KAAK,EAAE,KAAK,CAAC,CAAC;QACrE,WAAW,EAAE,OAAO,CAAC,WAAW;QAChC,UAAU,EAAE,OAAO,CAAC,UAAU;KAC/B,CAAC,CAAC;IACH,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,CAAC;QACjB,OAAO,CAAC,QAAQ,EAAE,EAAE,MAAM,EAAE,eAAe,EAAE,OAAO,EAAE,GAAG,EAAE,CAAC,CAAC;IAC/D,CAAC;IACD,MAAM,QAAQ,GAAG,QAAQ,CAAC,KAAK,CAAC;IAEhC,MAAM,CAAC,MAAM,CAAC,GAAG,MAAM,CAAC,CAAC,KAAK,CAW5B,CAAC,CAAC,IAAI,CAAC,GAAG,EAAE,UAAU,CAAC,kCAAkC,CAAC,EAC1D,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,EACb,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CACd,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC;IACjB,IAAI,CAAC,MAAM,CAAC,EAAE,EAAE,CAAC;QACf,OAAO,CAAC,MAAM,EAAE,EAAE,MAAM,EAAE,UAAU,EAAE,OAAO,EAAE,GAAG,EAAE,QAAQ,EAAE,CAAC,CAAC;IAClE,CAAC;IAED,OAAO,CAAC,MAAM,EAAE,EAAE,MAAM,EAAE,UAAU,EAAE,OAAO,EAAE,GAAG,EAAE,QAAQ,EAAE,CAAC,CAAC;AAClE,CAAC"}

package/dist/esm/funcs/llmEvalsImportAgentRun.d.ts

@@ -0,0 +1,19 @@
+import { SDKCore } from "../core.js";
+import { RequestOptions } from "../lib/sdks.js";
+import { ConnectionError, InvalidRequestError, RequestAbortedError, RequestTimeoutError, UnexpectedClientError } from "../models/errors/httpclienterrors.js";
+import { ResponseValidationError } from "../models/errors/responsevalidationerror.js";
+import { SDKError } from "../models/errors/sdkerror.js";
+import { SDKValidationError } from "../models/errors/sdkvalidationerror.js";
+import * as operations from "../models/operations/index.js";
+import { APIPromise } from "../types/async.js";
+import { Result } from "../types/fp.js";
+/**
+ * Import a finished coding-agent run
+ *
+ * @remarks
+ * Uploads one completed harness job — the run and its trials — directly, so results never round-trip through external storage the way the Hugging Face importer does. Every `task_id` must already exist in the target suite version's manifest; unknown ids are rejected rather than invented, because a typo would otherwise import as a task the suite does not have and read as a model that never covered the version. The run is recorded as `completed` and is not executed: this is a write, not a run request, so no workflow is started. Trials that cover every task of the version get `task_ids: null`, which is what lets the run rank as a complete sweep. Trajectories are NOT part of this payload — a full 91-task run measured 8.3 MB of them — so upload each one against `PUT /evals/runs/{run_id}/samples/{sample_index}/trajectory` afterwards. Pass `idempotency_key` and a retried upload returns the first run instead of doubling the trials behind a board row.
+ *
+ * If set, this operation will use {@link Security.bearerAuth} from the global security.
+ */
+export declare function llmEvalsImportAgentRun(client: SDKCore, request: operations.ImportAgentEvalRunRequest, options?: RequestOptions): APIPromise<Result<operations.ImportAgentEvalRunResponseBody, SDKError | ResponseValidationError | ConnectionError | RequestAbortedError | RequestTimeoutError | InvalidRequestError | UnexpectedClientError | SDKValidationError>>;
+//# sourceMappingURL=llmEvalsImportAgentRun.d.ts.map

package/dist/esm/funcs/llmEvalsImportAgentRun.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"llmEvalsImportAgentRun.d.ts","sourceRoot":"","sources":["../../../src/funcs/llmEvalsImportAgentRun.ts"],"names":[],"mappings":"AAIA,OAAO,EAAE,OAAO,EAAE,MAAM,YAAY,CAAC;AAMrC,OAAO,EAAE,cAAc,EAAE,MAAM,gBAAgB,CAAC;AAGhD,OAAO,EACL,eAAe,EACf,mBAAmB,EACnB,mBAAmB,EACnB,mBAAmB,EACnB,qBAAqB,EACtB,MAAM,sCAAsC,CAAC;AAC9C,OAAO,EAAE,uBAAuB,EAAE,MAAM,6CAA6C,CAAC;AACtF,OAAO,EAAE,QAAQ,EAAE,MAAM,8BAA8B,CAAC;AACxD,OAAO,EAAE,kBAAkB,EAAE,MAAM,wCAAwC,CAAC;AAC5E,OAAO,KAAK,UAAU,MAAM,+BAA+B,CAAC;AAC5D,OAAO,EAAW,UAAU,EAAE,MAAM,mBAAmB,CAAC;AACxD,OAAO,EAAE,MAAM,EAAE,MAAM,gBAAgB,CAAC;AAExC;;;;;;;GAOG;AACH,wBAAgB,sBAAsB,CACpC,MAAM,EAAE,OAAO,EACf,OAAO,EAAE,UAAU,CAAC,yBAAyB,EAC7C,OAAO,CAAC,EAAE,cAAc,GACvB,UAAU,CACX,MAAM,CACJ,UAAU,CAAC,8BAA8B,EACvC,QAAQ,GACR,uBAAuB,GACvB,eAAe,GACf,mBAAmB,GACnB,mBAAmB,GACnB,mBAAmB,GACnB,qBAAqB,GACrB,kBAAkB,CACrB,CACF,CAMA"}

package/dist/esm/funcs/llmEvalsImportAgentRun.js

@@ -0,0 +1,94 @@
+/*
+ * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
+ */
+import { encodeJSON, encodeSimple } from "../lib/encodings.js";
+import { matchStatusCode } from "../lib/http.js";
+import * as M from "../lib/matchers.js";
+import { compactMap } from "../lib/primitives.js";
+import { safeParse } from "../lib/schemas.js";
+import { extractSecurity, resolveGlobalSecurity } from "../lib/security.js";
+import { pathToFunc } from "../lib/url.js";
+import * as operations from "../models/operations/index.js";
+import { APIPromise } from "../types/async.js";
+/**
+ * Import a finished coding-agent run
+ *
+ * @remarks
+ * Uploads one completed harness job — the run and its trials — directly, so results never round-trip through external storage the way the Hugging Face importer does. Every `task_id` must already exist in the target suite version's manifest; unknown ids are rejected rather than invented, because a typo would otherwise import as a task the suite does not have and read as a model that never covered the version. The run is recorded as `completed` and is not executed: this is a write, not a run request, so no workflow is started. Trials that cover every task of the version get `task_ids: null`, which is what lets the run rank as a complete sweep. Trajectories are NOT part of this payload — a full 91-task run measured 8.3 MB of them — so upload each one against `PUT /evals/runs/{run_id}/samples/{sample_index}/trajectory` afterwards. Pass `idempotency_key` and a retried upload returns the first run instead of doubling the trials behind a board row.
+ *
+ * If set, this operation will use {@link Security.bearerAuth} from the global security.
+ */
+export function llmEvalsImportAgentRun(client, request, options) {
+    return new APIPromise($do(client, request, options));
+}
+async function $do(client, request, options) {
+    const parsed = safeParse(request, (value) => operations.ImportAgentEvalRunRequest$outboundSchema.parse(value), "Input validation failed");
+    if (!parsed.ok) {
+        return [parsed, { status: "invalid" }];
+    }
+    const payload = parsed.value;
+    const body = encodeJSON("body", payload.ImportAgentRunRequest, {
+        explode: true,
+    });
+    const path = pathToFunc("/api/v1/llm/evals/imports/agent-run")();
+    const headers = new Headers(compactMap({
+        "Content-Type": "application/json",
+        Accept: "application/json",
+        "X-On-Behalf-Of": encodeSimple("X-On-Behalf-Of", payload["X-On-Behalf-Of"], { explode: false, charEncoding: "none" }),
+    }));
+    const secConfig = await extractSecurity(client._options.bearerAuth);
+    const securityInput = secConfig == null ? {} : { bearerAuth: secConfig };
+    const requestSecurity = resolveGlobalSecurity(securityInput, [0]);
+    const context = {
+        options: client._options,
+        baseURL: options?.serverURL ?? client._baseURL ?? "",
+        operationID: "importAgentEvalRun",
+        oAuth2Scopes: null,
+        resolvedSecurity: requestSecurity,
+        securitySource: client._options.bearerAuth,
+        retryConfig: options?.retries
+            || client._options.retryConfig
+            || {
+                strategy: "backoff",
+                backoff: {
+                    initialInterval: 500,
+                    maxInterval: 5000,
+                    exponent: 1.5,
+                    maxElapsedTime: 15000,
+                },
+                retryConnectionErrors: true,
+            }
+            || { strategy: "none" },
+        retryCodes: options?.retryCodes || ["429", "500", "502", "503", "504"],
+    };
+    const requestRes = client._createRequest(context, {
+        security: requestSecurity,
+        method: "POST",
+        baseURL: options?.serverURL,
+        path: path,
+        headers: headers,
+        body: body,
+        userAgent: client._options.userAgent,
+        timeoutMs: options?.timeoutMs || client._options.timeoutMs || -1,
+    }, options);
+    if (!requestRes.ok) {
+        return [requestRes, { status: "invalid" }];
+    }
+    const req = requestRes.value;
+    const doResult = await client._do(req, {
+        context,
+        isErrorStatusCode: (statusCode) => matchStatusCode({ status: statusCode }, ["4XX", "5XX"]),
+        retryConfig: context.retryConfig,
+        retryCodes: context.retryCodes,
+    });
+    if (!doResult.ok) {
+        return [doResult, { status: "request-error", request: req }];
+    }
+    const response = doResult.value;
+    const [result] = await M.match(M.json(200, operations.ImportAgentEvalRunResponseBody$inboundSchema), M.fail("4XX"), M.fail("5XX"))(response, req);
+    if (!result.ok) {
+        return [result, { status: "complete", request: req, response }];
+    }
+    return [result, { status: "complete", request: req, response }];
+}
+//# sourceMappingURL=llmEvalsImportAgentRun.js.map

package/dist/esm/funcs/llmEvalsImportAgentRun.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"llmEvalsImportAgentRun.js","sourceRoot":"","sources":["../../../src/funcs/llmEvalsImportAgentRun.ts"],"names":[],"mappings":"AAAA;;GAEG;AAGH,OAAO,EAAE,UAAU,EAAE,YAAY,EAAE,MAAM,qBAAqB,CAAC;AAC/D,OAAO,EAAE,eAAe,EAAE,MAAM,gBAAgB,CAAC;AACjD,OAAO,KAAK,CAAC,MAAM,oBAAoB,CAAC;AACxC,OAAO,EAAE,UAAU,EAAE,MAAM,sBAAsB,CAAC;AAClD,OAAO,EAAE,SAAS,EAAE,MAAM,mBAAmB,CAAC;AAE9C,OAAO,EAAE,eAAe,EAAE,qBAAqB,EAAE,MAAM,oBAAoB,CAAC;AAC5E,OAAO,EAAE,UAAU,EAAE,MAAM,eAAe,CAAC;AAW3C,OAAO,KAAK,UAAU,MAAM,+BAA+B,CAAC;AAC5D,OAAO,EAAW,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAGxD;;;;;;;GAOG;AACH,MAAM,UAAU,sBAAsB,CACpC,MAAe,EACf,OAA6C,EAC7C,OAAwB;IAcxB,OAAO,IAAI,UAAU,CAAC,GAAG,CACvB,MAAM,EACN,OAAO,EACP,OAAO,CACR,CAAC,CAAC;AACL,CAAC;AAED,KAAK,UAAU,GAAG,CAChB,MAAe,EACf,OAA6C,EAC7C,OAAwB;IAiBxB,MAAM,MAAM,GAAG,SAAS,CACtB,OAAO,EACP,CAAC,KAAK,EAAE,EAAE,CAAC,UAAU,CAAC,wCAAwC,CAAC,KAAK,CAAC,KAAK,CAAC,EAC3E,yBAAyB,CAC1B,CAAC;IACF,IAAI,CAAC,MAAM,CAAC,EAAE,EAAE,CAAC;QACf,OAAO,CAAC,MAAM,EAAE,EAAE,MAAM,EAAE,SAAS,EAAE,CAAC,CAAC;IACzC,CAAC;IACD,MAAM,OAAO,GAAG,MAAM,CAAC,KAAK,CAAC;IAC7B,MAAM,IAAI,GAAG,UAAU,CAAC,MAAM,EAAE,OAAO,CAAC,qBAAqB,EAAE;QAC7D,OAAO,EAAE,IAAI;KACd,CAAC,CAAC;IAEH,MAAM,IAAI,GAAG,UAAU,CAAC,qCAAqC,CAAC,EAAE,CAAC;IAEjE,MAAM,OAAO,GAAG,IAAI,OAAO,CAAC,UAAU,CAAC;QACrC,cAAc,EAAE,kBAAkB;QAClC,MAAM,EAAE,kBAAkB;QAC1B,gBAAgB,EAAE,YAAY,CAC5B,gBAAgB,EAChB,OAAO,CAAC,gBAAgB,CAAC,EACzB,EAAE,OAAO,EAAE,KAAK,EAAE,YAAY,EAAE,MAAM,EAAE,CACzC;KACF,CAAC,CAAC,CAAC;IAEJ,MAAM,SAAS,GAAG,MAAM,eAAe,CAAC,MAAM,CAAC,QAAQ,CAAC,UAAU,CAAC,CAAC;IACpE,MAAM,aAAa,GAAG,SAAS,IAAI,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,SAAS,EAAE,CAAC;IACzE,MAAM,eAAe,GAAG,qBAAqB,CAAC,aAAa,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;IAElE,MAAM,OAAO,GAAG;QACd,OAAO,EAAE,MAAM,CAAC,QAAQ;QACxB,OAAO,EAAE,OAAO,EAAE,SAAS,IAAI,MAAM,CAAC,QAAQ,IAAI,EAAE;QACpD,WAAW,EAAE,oBAAoB;QACjC,YAAY,EAAE,IAAI;QAElB,gBAAgB,EAAE,eAAe;QAEjC,cAAc,EAAE,MAAM,CAAC,QAAQ,CAAC,UAAU;QAC1C,WAAW,EAAE,OAAO,EAAE,OAAO;eACxB,MAAM,CAAC,QAAQ,CAAC,WAAW;eAC3B;gBACD,QAAQ,EAAE,SAAS;gBACnB,OAAO,EAAE;oBACP,eAAe,EAAE,GAAG;oBACpB,WAAW,EAAE,IAAI;oBACjB,QAAQ,EAAE,GAAG;oBACb,cAAc,EAAE,KAAK;iBACtB;gBACD,qBAAqB,EAAE,IAAI;aAC5B;eACE,EAAE,QAAQ,EAAE,MAAM,EAAE;QACzB,UAAU,EAAE,OAAO,EAAE,UAAU,IAAI,CAAC,KAAK,EAAE,KAAK,EAAE,KAAK,EAAE,KAAK,EAAE,KAAK,CAAC;KACvE,CAAC;IAEF,MAAM,UAAU,GAAG,MAAM,CAAC,cAAc,CAAC,OAAO,EAAE;QAChD,QAAQ,EAAE,eAAe;QACzB,MAAM,EAAE,MAAM;QACd,OAAO,EAAE,OAAO,EAAE,SAAS;QAC3B,IAAI,EAAE,IAAI;QACV,OAAO,EAAE,OAAO;QAChB,IAAI,EAAE,IAAI;QACV,SAAS,EAAE,MAAM,CAAC,QAAQ,CAAC,SAAS;QACpC,SAAS,EAAE,OAAO,EAAE,SAAS,IAAI,MAAM,CAAC,QAAQ,CAAC,SAAS,IAAI,CAAC,CAAC;KACjE,EAAE,OAAO,CAAC,CAAC;IACZ,IAAI,CAAC,UAAU,CAAC,EAAE,EAAE,CAAC;QACnB,OAAO,CAAC,UAAU,EAAE,EAAE,MAAM,EAAE,SAAS,EAAE,CAAC,CAAC;IAC7C,CAAC;IACD,MAAM,GAAG,GAAG,UAAU,CAAC,KAAK,CAAC;IAE7B,MAAM,QAAQ,GAAG,MAAM,MAAM,CAAC,GAAG,CAAC,GAAG,EAAE;QACrC,OAAO;QACP,iBAAiB,EAAE,CAAC,UAAkB,EAAE,EAAE,CACxC,eAAe,CAAC,EAAE,MAAM,EAAE,UAAU,EAAc,EAAE,CAAC,KAAK,EAAE,KAAK,CAAC,CAAC;QACrE,WAAW,EAAE,OAAO,CAAC,WAAW;QAChC,UAAU,EAAE,OAAO,CAAC,UAAU;KAC/B,CAAC,CAAC;IACH,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,CAAC;QACjB,OAAO,CAAC,QAAQ,EAAE,EAAE,MAAM,EAAE,eAAe,EAAE,OAAO,EAAE,GAAG,EAAE,CAAC,CAAC;IAC/D,CAAC;IACD,MAAM,QAAQ,GAAG,QAAQ,CAAC,KAAK,CAAC;IAEhC,MAAM,CAAC,MAAM,CAAC,GAAG,MAAM,CAAC,CAAC,KAAK,CAW5B,CAAC,CAAC,IAAI,CAAC,GAAG,EAAE,UAAU,CAAC,4CAA4C,CAAC,EACpE,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,EACb,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CACd,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC;IACjB,IAAI,CAAC,MAAM,CAAC,EAAE,EAAE,CAAC;QACf,OAAO,CAAC,MAAM,EAAE,EAAE,MAAM,EAAE,UAAU,EAAE,OAAO,EAAE,GAAG,EAAE,QAAQ,EAAE,CAAC,CAAC;IAClE,CAAC;IAED,OAAO,CAAC,MAAM,EAAE,EAAE,MAAM,EAAE,UAAU,EAAE,OAAO,EAAE,GAAG,EAAE,QAAQ,EAAE,CAAC,CAAC;AAClE,CAAC"}

package/dist/esm/funcs/llmEvalsUploadSampleTrajectory.d.ts

@@ -0,0 +1,20 @@
+import { SDKCore } from "../core.js";
+import { RequestOptions } from "../lib/sdks.js";
+import * as components from "../models/components/index.js";
+import { ConnectionError, InvalidRequestError, RequestAbortedError, RequestTimeoutError, UnexpectedClientError } from "../models/errors/httpclienterrors.js";
+import { ResponseValidationError } from "../models/errors/responsevalidationerror.js";
+import { SDKError } from "../models/errors/sdkerror.js";
+import { SDKValidationError } from "../models/errors/sdkvalidationerror.js";
+import * as operations from "../models/operations/index.js";
+import { APIPromise } from "../types/async.js";
+import { Result } from "../types/fp.js";
+/**
+ * Upload an eval sample trajectory
+ *
+ * @remarks
+ * Stores one coding-agent trial's session, in whatever interchange format the harness produced (record it in `schema_version`, e.g. `ATIF-v1.7`). `steps` is stored verbatim and never validated, so a harness format change needs no gateway release. Separate from the run import because a trajectory averages ~92 KB and a whole run's worth would not fit in one request. Upserts, so re-sending after a timeout replaces the trajectory rather than failing. Sample indices are unique per (task, model), not per run, so pass `task_id` and/or `model` when the run covers more than one — an ambiguous index returns 400 rather than writing to the wrong trial. Rejects a payload whose `steps` and `logs` together exceed 16 MiB with 413.
+ *
+ * If set, this operation will use {@link Security.bearerAuth} from the global security.
+ */
+export declare function llmEvalsUploadSampleTrajectory(client: SDKCore, request: operations.UploadEvalSampleTrajectoryRequest, options?: RequestOptions): APIPromise<Result<components.EvalSampleTrajectoryUploadedObject, SDKError | ResponseValidationError | ConnectionError | RequestAbortedError | RequestTimeoutError | InvalidRequestError | UnexpectedClientError | SDKValidationError>>;
+//# sourceMappingURL=llmEvalsUploadSampleTrajectory.d.ts.map

package/dist/esm/funcs/llmEvalsUploadSampleTrajectory.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"llmEvalsUploadSampleTrajectory.d.ts","sourceRoot":"","sources":["../../../src/funcs/llmEvalsUploadSampleTrajectory.ts"],"names":[],"mappings":"AAIA,OAAO,EAAE,OAAO,EAAE,MAAM,YAAY,CAAC;AAMrC,OAAO,EAAE,cAAc,EAAE,MAAM,gBAAgB,CAAC;AAGhD,OAAO,KAAK,UAAU,MAAM,+BAA+B,CAAC;AAC5D,OAAO,EACL,eAAe,EACf,mBAAmB,EACnB,mBAAmB,EACnB,mBAAmB,EACnB,qBAAqB,EACtB,MAAM,sCAAsC,CAAC;AAC9C,OAAO,EAAE,uBAAuB,EAAE,MAAM,6CAA6C,CAAC;AACtF,OAAO,EAAE,QAAQ,EAAE,MAAM,8BAA8B,CAAC;AACxD,OAAO,EAAE,kBAAkB,EAAE,MAAM,wCAAwC,CAAC;AAC5E,OAAO,KAAK,UAAU,MAAM,+BAA+B,CAAC;AAC5D,OAAO,EAAW,UAAU,EAAE,MAAM,mBAAmB,CAAC;AACxD,OAAO,EAAE,MAAM,EAAE,MAAM,gBAAgB,CAAC;AAExC;;;;;;;GAOG;AACH,wBAAgB,8BAA8B,CAC5C,MAAM,EAAE,OAAO,EACf,OAAO,EAAE,UAAU,CAAC,iCAAiC,EACrD,OAAO,CAAC,EAAE,cAAc,GACvB,UAAU,CACX,MAAM,CACJ,UAAU,CAAC,kCAAkC,EAC3C,QAAQ,GACR,uBAAuB,GACvB,eAAe,GACf,mBAAmB,GACnB,mBAAmB,GACnB,mBAAmB,GACnB,qBAAqB,GACrB,kBAAkB,CACrB,CACF,CAMA"}

package/dist/esm/funcs/llmEvalsUploadSampleTrajectory.js

@@ -0,0 +1,110 @@
+/*
+ * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
+ */
+import { encodeFormQuery, encodeJSON, encodeSimple } from "../lib/encodings.js";
+import { matchStatusCode } from "../lib/http.js";
+import * as M from "../lib/matchers.js";
+import { compactMap } from "../lib/primitives.js";
+import { safeParse } from "../lib/schemas.js";
+import { extractSecurity, resolveGlobalSecurity } from "../lib/security.js";
+import { pathToFunc } from "../lib/url.js";
+import * as components from "../models/components/index.js";
+import * as operations from "../models/operations/index.js";
+import { APIPromise } from "../types/async.js";
+/**
+ * Upload an eval sample trajectory
+ *
+ * @remarks
+ * Stores one coding-agent trial's session, in whatever interchange format the harness produced (record it in `schema_version`, e.g. `ATIF-v1.7`). `steps` is stored verbatim and never validated, so a harness format change needs no gateway release. Separate from the run import because a trajectory averages ~92 KB and a whole run's worth would not fit in one request. Upserts, so re-sending after a timeout replaces the trajectory rather than failing. Sample indices are unique per (task, model), not per run, so pass `task_id` and/or `model` when the run covers more than one — an ambiguous index returns 400 rather than writing to the wrong trial. Rejects a payload whose `steps` and `logs` together exceed 16 MiB with 413.
+ *
+ * If set, this operation will use {@link Security.bearerAuth} from the global security.
+ */
+export function llmEvalsUploadSampleTrajectory(client, request, options) {
+    return new APIPromise($do(client, request, options));
+}
+async function $do(client, request, options) {
+    const parsed = safeParse(request, (value) => operations.UploadEvalSampleTrajectoryRequest$outboundSchema.parse(value), "Input validation failed");
+    if (!parsed.ok) {
+        return [parsed, { status: "invalid" }];
+    }
+    const payload = parsed.value;
+    const body = encodeJSON("body", payload.UploadEvalSampleTrajectoryRequest, {
+        explode: true,
+    });
+    const pathParams = {
+        run_id: encodeSimple("run_id", payload.run_id, {
+            explode: false,
+            charEncoding: "percent",
+        }),
+        sample_index: encodeSimple("sample_index", payload.sample_index, {
+            explode: false,
+            charEncoding: "percent",
+        }),
+    };
+    const path = pathToFunc("/api/v1/llm/evals/runs/{run_id}/samples/{sample_index}/trajectory")(pathParams);
+    const query = encodeFormQuery({
+        "model": payload.model,
+        "task_id": payload.task_id,
+    });
+    const headers = new Headers(compactMap({
+        "Content-Type": "application/json",
+        Accept: "application/json",
+        "X-On-Behalf-Of": encodeSimple("X-On-Behalf-Of", payload["X-On-Behalf-Of"], { explode: false, charEncoding: "none" }),
+    }));
+    const secConfig = await extractSecurity(client._options.bearerAuth);
+    const securityInput = secConfig == null ? {} : { bearerAuth: secConfig };
+    const requestSecurity = resolveGlobalSecurity(securityInput, [0]);
+    const context = {
+        options: client._options,
+        baseURL: options?.serverURL ?? client._baseURL ?? "",
+        operationID: "uploadEvalSampleTrajectory",
+        oAuth2Scopes: null,
+        resolvedSecurity: requestSecurity,
+        securitySource: client._options.bearerAuth,
+        retryConfig: options?.retries
+            || client._options.retryConfig
+            || {
+                strategy: "backoff",
+                backoff: {
+                    initialInterval: 500,
+                    maxInterval: 5000,
+                    exponent: 1.5,
+                    maxElapsedTime: 15000,
+                },
+                retryConnectionErrors: true,
+            }
+            || { strategy: "none" },
+        retryCodes: options?.retryCodes || ["429", "500", "502", "503", "504"],
+    };
+    const requestRes = client._createRequest(context, {
+        security: requestSecurity,
+        method: "PUT",
+        baseURL: options?.serverURL,
+        path: path,
+        headers: headers,
+        query: query,
+        body: body,
+        userAgent: client._options.userAgent,
+        timeoutMs: options?.timeoutMs || client._options.timeoutMs || -1,
+    }, options);
+    if (!requestRes.ok) {
+        return [requestRes, { status: "invalid" }];
+    }
+    const req = requestRes.value;
+    const doResult = await client._do(req, {
+        context,
+        isErrorStatusCode: (statusCode) => matchStatusCode({ status: statusCode }, ["4XX", "5XX"]),
+        retryConfig: context.retryConfig,
+        retryCodes: context.retryCodes,
+    });
+    if (!doResult.ok) {
+        return [doResult, { status: "request-error", request: req }];
+    }
+    const response = doResult.value;
+    const [result] = await M.match(M.json(200, components.EvalSampleTrajectoryUploadedObject$inboundSchema), M.fail("4XX"), M.fail("5XX"))(response, req);
+    if (!result.ok) {
+        return [result, { status: "complete", request: req, response }];
+    }
+    return [result, { status: "complete", request: req, response }];
+}
+//# sourceMappingURL=llmEvalsUploadSampleTrajectory.js.map

package/dist/esm/funcs/llmEvalsUploadSampleTrajectory.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"llmEvalsUploadSampleTrajectory.js","sourceRoot":"","sources":["../../../src/funcs/llmEvalsUploadSampleTrajectory.ts"],"names":[],"mappings":"AAAA;;GAEG;AAGH,OAAO,EAAE,eAAe,EAAE,UAAU,EAAE,YAAY,EAAE,MAAM,qBAAqB,CAAC;AAChF,OAAO,EAAE,eAAe,EAAE,MAAM,gBAAgB,CAAC;AACjD,OAAO,KAAK,CAAC,MAAM,oBAAoB,CAAC;AACxC,OAAO,EAAE,UAAU,EAAE,MAAM,sBAAsB,CAAC;AAClD,OAAO,EAAE,SAAS,EAAE,MAAM,mBAAmB,CAAC;AAE9C,OAAO,EAAE,eAAe,EAAE,qBAAqB,EAAE,MAAM,oBAAoB,CAAC;AAC5E,OAAO,EAAE,UAAU,EAAE,MAAM,eAAe,CAAC;AAC3C,OAAO,KAAK,UAAU,MAAM,+BAA+B,CAAC;AAW5D,OAAO,KAAK,UAAU,MAAM,+BAA+B,CAAC;AAC5D,OAAO,EAAW,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAGxD;;;;;;;GAOG;AACH,MAAM,UAAU,8BAA8B,CAC5C,MAAe,EACf,OAAqD,EACrD,OAAwB;IAcxB,OAAO,IAAI,UAAU,CAAC,GAAG,CACvB,MAAM,EACN,OAAO,EACP,OAAO,CACR,CAAC,CAAC;AACL,CAAC;AAED,KAAK,UAAU,GAAG,CAChB,MAAe,EACf,OAAqD,EACrD,OAAwB;IAiBxB,MAAM,MAAM,GAAG,SAAS,CACtB,OAAO,EACP,CAAC,KAAK,EAAE,EAAE,CACR,UAAU,CAAC,gDAAgD,CAAC,KAAK,CAAC,KAAK,CAAC,EAC1E,yBAAyB,CAC1B,CAAC;IACF,IAAI,CAAC,MAAM,CAAC,EAAE,EAAE,CAAC;QACf,OAAO,CAAC,MAAM,EAAE,EAAE,MAAM,EAAE,SAAS,EAAE,CAAC,CAAC;IACzC,CAAC;IACD,MAAM,OAAO,GAAG,MAAM,CAAC,KAAK,CAAC;IAC7B,MAAM,IAAI,GAAG,UAAU,CAAC,MAAM,EAAE,OAAO,CAAC,iCAAiC,EAAE;QACzE,OAAO,EAAE,IAAI;KACd,CAAC,CAAC;IAEH,MAAM,UAAU,GAAG;QACjB,MAAM,EAAE,YAAY,CAAC,QAAQ,EAAE,OAAO,CAAC,MAAM,EAAE;YAC7C,OAAO,EAAE,KAAK;YACd,YAAY,EAAE,SAAS;SACxB,CAAC;QACF,YAAY,EAAE,YAAY,CAAC,cAAc,EAAE,OAAO,CAAC,YAAY,EAAE;YAC/D,OAAO,EAAE,KAAK;YACd,YAAY,EAAE,SAAS;SACxB,CAAC;KACH,CAAC;IACF,MAAM,IAAI,GAAG,UAAU,CACrB,mEAAmE,CACpE,CAAC,UAAU,CAAC,CAAC;IAEd,MAAM,KAAK,GAAG,eAAe,CAAC;QAC5B,OAAO,EAAE,OAAO,CAAC,KAAK;QACtB,SAAS,EAAE,OAAO,CAAC,OAAO;KAC3B,CAAC,CAAC;IAEH,MAAM,OAAO,GAAG,IAAI,OAAO,CAAC,UAAU,CAAC;QACrC,cAAc,EAAE,kBAAkB;QAClC,MAAM,EAAE,kBAAkB;QAC1B,gBAAgB,EAAE,YAAY,CAC5B,gBAAgB,EAChB,OAAO,CAAC,gBAAgB,CAAC,EACzB,EAAE,OAAO,EAAE,KAAK,EAAE,YAAY,EAAE,MAAM,EAAE,CACzC;KACF,CAAC,CAAC,CAAC;IAEJ,MAAM,SAAS,GAAG,MAAM,eAAe,CAAC,MAAM,CAAC,QAAQ,CAAC,UAAU,CAAC,CAAC;IACpE,MAAM,aAAa,GAAG,SAAS,IAAI,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,SAAS,EAAE,CAAC;IACzE,MAAM,eAAe,GAAG,qBAAqB,CAAC,aAAa,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;IAElE,MAAM,OAAO,GAAG;QACd,OAAO,EAAE,MAAM,CAAC,QAAQ;QACxB,OAAO,EAAE,OAAO,EAAE,SAAS,IAAI,MAAM,CAAC,QAAQ,IAAI,EAAE;QACpD,WAAW,EAAE,4BAA4B;QACzC,YAAY,EAAE,IAAI;QAElB,gBAAgB,EAAE,eAAe;QAEjC,cAAc,EAAE,MAAM,CAAC,QAAQ,CAAC,UAAU;QAC1C,WAAW,EAAE,OAAO,EAAE,OAAO;eACxB,MAAM,CAAC,QAAQ,CAAC,WAAW;eAC3B;gBACD,QAAQ,EAAE,SAAS;gBACnB,OAAO,EAAE;oBACP,eAAe,EAAE,GAAG;oBACpB,WAAW,EAAE,IAAI;oBACjB,QAAQ,EAAE,GAAG;oBACb,cAAc,EAAE,KAAK;iBACtB;gBACD,qBAAqB,EAAE,IAAI;aAC5B;eACE,EAAE,QAAQ,EAAE,MAAM,EAAE;QACzB,UAAU,EAAE,OAAO,EAAE,UAAU,IAAI,CAAC,KAAK,EAAE,KAAK,EAAE,KAAK,EAAE,KAAK,EAAE,KAAK,CAAC;KACvE,CAAC;IAEF,MAAM,UAAU,GAAG,MAAM,CAAC,cAAc,CAAC,OAAO,EAAE;QAChD,QAAQ,EAAE,eAAe;QACzB,MAAM,EAAE,KAAK;QACb,OAAO,EAAE,OAAO,EAAE,SAAS;QAC3B,IAAI,EAAE,IAAI;QACV,OAAO,EAAE,OAAO;QAChB,KAAK,EAAE,KAAK;QACZ,IAAI,EAAE,IAAI;QACV,SAAS,EAAE,MAAM,CAAC,QAAQ,CAAC,SAAS;QACpC,SAAS,EAAE,OAAO,EAAE,SAAS,IAAI,MAAM,CAAC,QAAQ,CAAC,SAAS,IAAI,CAAC,CAAC;KACjE,EAAE,OAAO,CAAC,CAAC;IACZ,IAAI,CAAC,UAAU,CAAC,EAAE,EAAE,CAAC;QACnB,OAAO,CAAC,UAAU,EAAE,EAAE,MAAM,EAAE,SAAS,EAAE,CAAC,CAAC;IAC7C,CAAC;IACD,MAAM,GAAG,GAAG,UAAU,CAAC,KAAK,CAAC;IAE7B,MAAM,QAAQ,GAAG,MAAM,MAAM,CAAC,GAAG,CAAC,GAAG,EAAE;QACrC,OAAO;QACP,iBAAiB,EAAE,CAAC,UAAkB,EAAE,EAAE,CACxC,eAAe,CAAC,EAAE,MAAM,EAAE,UAAU,EAAc,EAAE,CAAC,KAAK,EAAE,KAAK,CAAC,CAAC;QACrE,WAAW,EAAE,OAAO,CAAC,WAAW;QAChC,UAAU,EAAE,OAAO,CAAC,UAAU;KAC/B,CAAC,CAAC;IACH,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,CAAC;QACjB,OAAO,CAAC,QAAQ,EAAE,EAAE,MAAM,EAAE,eAAe,EAAE,OAAO,EAAE,GAAG,EAAE,CAAC,CAAC;IAC/D,CAAC;IACD,MAAM,QAAQ,GAAG,QAAQ,CAAC,KAAK,CAAC;IAEhC,MAAM,CAAC,MAAM,CAAC,GAAG,MAAM,CAAC,CAAC,KAAK,CAW5B,CAAC,CAAC,IAAI,CAAC,GAAG,EAAE,UAAU,CAAC,gDAAgD,CAAC,EACxE,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,EACb,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CACd,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC;IACjB,IAAI,CAAC,MAAM,CAAC,EAAE,EAAE,CAAC;QACf,OAAO,CAAC,MAAM,EAAE,EAAE,MAAM,EAAE,UAAU,EAAE,OAAO,EAAE,GAAG,EAAE,QAAQ,EAAE,CAAC,CAAC;IAClE,CAAC;IAED,OAAO,CAAC,MAAM,EAAE,EAAE,MAAM,EAAE,UAAU,EAAE,OAAO,EAAE,GAAG,EAAE,QAAQ,EAAE,CAAC,CAAC;AAClE,CAAC"}

package/dist/esm/lib/config.d.ts

@@ -31,8 +31,8 @@ export declare function serverURLFromOptions(options: SDKOptions): URL | null;
 export declare const SDK_METADATA: {
     readonly language: "typescript";
     readonly openapiDocVersion: "1.1.0";
-    readonly sdkVersion: "0.54.23";
+    readonly sdkVersion: "0.54.25";
     readonly genVersion: "2.935.1";
-    readonly userAgent: "speakeasy-sdk/typescript 0.54.23 2.935.1 1.1.0 @meetkai/mka1";
+    readonly userAgent: "speakeasy-sdk/typescript 0.54.25 2.935.1 1.1.0 @meetkai/mka1";
 };
 //# sourceMappingURL=config.d.ts.map

package/dist/esm/lib/config.js

@@ -31,8 +31,8 @@ export function serverURLFromOptions(options) {
 export const SDK_METADATA = {
     language: "typescript",
     openapiDocVersion: "1.1.0",
-    sdkVersion: "0.54.23",
+    sdkVersion: "0.54.25",
     genVersion: "2.935.1",
-    userAgent: "speakeasy-sdk/typescript 0.54.23 2.935.1 1.1.0 @meetkai/mka1",
+    userAgent: "speakeasy-sdk/typescript 0.54.25 2.935.1 1.1.0 @meetkai/mka1",
 };
 //# sourceMappingURL=config.js.map

package/dist/esm/mcp-server/mcp-server.js

@@ -16,7 +16,7 @@ const routes = buildRouteMap({
 export const app = buildApplication(routes, {
     name: "mcp",
     versionInfo: {
-        currentVersion: "0.54.23",
+        currentVersion: "0.54.25",
     },
 });
 run(app, process.argv.slice(2), buildContext(process));

package/dist/esm/mcp-server/server.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"server.d.ts","sourceRoot":"","sources":["../../../src/mcp-server/server.ts"],"names":[],"mappings":"AAIA,OAAO,EAAE,SAAS,EAAE,MAAM,yCAAyC,CAAC;AAEpE,OAAO,EAAE,UAAU,EAAE,MAAM,kBAAkB,CAAC;AAC9C,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,qBAAqB,CAAC;AAMzD,OAAO,EAAE,QAAQ,EAAE,MAAM,aAAa,CAAC;AAiYvC,wBAAgB,eAAe,CAAC,IAAI,EAAE;IACpC,MAAM,EAAE,aAAa,CAAC;IACtB,YAAY,CAAC,EAAE,MAAM,EAAE,GAAG,SAAS,CAAC;IACpC,MAAM,CAAC,EAAE,QAAQ,EAAE,GAAG,SAAS,CAAC;IAChC,SAAS,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAC/B,UAAU,CAAC,EAAE,UAAU,CAAC,YAAY,CAAC,GAAG,SAAS,CAAC;IAClD,SAAS,CAAC,EAAE,UAAU,CAAC,WAAW,CAAC,GAAG,SAAS,CAAC;CACjD,aA6aA"}
+{"version":3,"file":"server.d.ts","sourceRoot":"","sources":["../../../src/mcp-server/server.ts"],"names":[],"mappings":"AAIA,OAAO,EAAE,SAAS,EAAE,MAAM,yCAAyC,CAAC;AAEpE,OAAO,EAAE,UAAU,EAAE,MAAM,kBAAkB,CAAC;AAC9C,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,qBAAqB,CAAC;AAMzD,OAAO,EAAE,QAAQ,EAAE,MAAM,aAAa,CAAC;AAuYvC,wBAAgB,eAAe,CAAC,IAAI,EAAE;IACpC,MAAM,EAAE,aAAa,CAAC;IACtB,YAAY,CAAC,EAAE,MAAM,EAAE,GAAG,SAAS,CAAC;IACpC,MAAM,CAAC,EAAE,QAAQ,EAAE,GAAG,SAAS,CAAC;IAChC,SAAS,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAC/B,UAAU,CAAC,EAAE,UAAU,CAAC,YAAY,CAAC,GAAG,SAAS,CAAC;IAClD,SAAS,CAAC,EAAE,UAAU,CAAC,WAAW,CAAC,GAAG,SAAS,CAAC;CACjD,aAmbA"}

package/dist/esm/mcp-server/server.js

@@ -164,6 +164,7 @@ import { tool$llmConversationsListItems } from "./tools/llmConversationsListItem
 import { tool$llmConversationsUpdate } from "./tools/llmConversationsUpdate.js";
 import { tool$llmEmbeddingsEmbed } from "./tools/llmEmbeddingsEmbed.js";
 import { tool$llmEmbeddingsListModels } from "./tools/llmEmbeddingsListModels.js";
+import { tool$llmEvalsAppendAgentTrials } from "./tools/llmEvalsAppendAgentTrials.js";
 import { tool$llmEvalsCancelRun } from "./tools/llmEvalsCancelRun.js";
 import { tool$llmEvalsCreateRun } from "./tools/llmEvalsCreateRun.js";
 import { tool$llmEvalsCreateSchedule } from "./tools/llmEvalsCreateSchedule.js";
@@ -174,10 +175,14 @@ import { tool$llmEvalsDeleteSchedule } from "./tools/llmEvalsDeleteSchedule.js";
 import { tool$llmEvalsDeleteSuite } from "./tools/llmEvalsDeleteSuite.js";
 import { tool$llmEvalsGetArtifacts } from "./tools/llmEvalsGetArtifacts.js";
 import { tool$llmEvalsGetRun } from "./tools/llmEvalsGetRun.js";
+import { tool$llmEvalsGetRunOverview } from "./tools/llmEvalsGetRunOverview.js";
 import { tool$llmEvalsGetSampleAudio } from "./tools/llmEvalsGetSampleAudio.js";
+import { tool$llmEvalsGetSampleTrajectory } from "./tools/llmEvalsGetSampleTrajectory.js";
 import { tool$llmEvalsGetSchedule } from "./tools/llmEvalsGetSchedule.js";
 import { tool$llmEvalsGetSuite } from "./tools/llmEvalsGetSuite.js";
+import { tool$llmEvalsGetSuiteLeaderboard } from "./tools/llmEvalsGetSuiteLeaderboard.js";
 import { tool$llmEvalsGetSuiteVersion } from "./tools/llmEvalsGetSuiteVersion.js";
+import { tool$llmEvalsImportAgentRun } from "./tools/llmEvalsImportAgentRun.js";
 import { tool$llmEvalsImportHistoricalResults } from "./tools/llmEvalsImportHistoricalResults.js";
 import { tool$llmEvalsListRuns } from "./tools/llmEvalsListRuns.js";
 import { tool$llmEvalsListSamples } from "./tools/llmEvalsListSamples.js";
@@ -190,6 +195,7 @@ import { tool$llmEvalsRerunFailedSamples } from "./tools/llmEvalsRerunFailedSamp
 import { tool$llmEvalsRetryFailedRun } from "./tools/llmEvalsRetryFailedRun.js";
 import { tool$llmEvalsTriggerSchedule } from "./tools/llmEvalsTriggerSchedule.js";
 import { tool$llmEvalsUpdateSchedule } from "./tools/llmEvalsUpdateSchedule.js";
+import { tool$llmEvalsUploadSampleTrajectory } from "./tools/llmEvalsUploadSampleTrajectory.js";
 import { tool$llmExtractCreateSchema } from "./tools/llmExtractCreateSchema.js";
 import { tool$llmExtractDeleteSchema } from "./tools/llmExtractDeleteSchema.js";
 import { tool$llmExtractExtract } from "./tools/llmExtractExtract.js";
@@ -383,7 +389,7 @@ import { tool$usageSandbox } from "./tools/usageSandbox.js";
 export function createMCPServer(deps) {
     const server = new McpServer({
         name: "SDK",
-        version: "0.54.23",
+        version: "0.54.25",
     });
     const client = new SDKCore({
         bearerAuth: deps.bearerAuth,
@@ -569,6 +575,7 @@ export function createMCPServer(deps) {
     tool(tool$llmEvalsListSuiteVersions);
     tool(tool$llmEvalsGetSuiteVersion);
     tool(tool$llmEvalsListSuiteLeaderboardRuns);
+    tool(tool$llmEvalsGetSuiteLeaderboard);
     tool(tool$llmEvalsCreateRun);
     tool(tool$llmEvalsListRuns);
     tool(tool$llmEvalsGetRun);
@@ -576,9 +583,14 @@ export function createMCPServer(deps) {
     tool(tool$llmEvalsCancelRun);
     tool(tool$llmEvalsRerunFailedSamples);
     tool(tool$llmEvalsRetryFailedRun);
+    tool(tool$llmEvalsGetRunOverview);
     tool(tool$llmEvalsListSamples);
     tool(tool$llmEvalsGetSampleAudio);
+    tool(tool$llmEvalsGetSampleTrajectory);
+    tool(tool$llmEvalsUploadSampleTrajectory);
     tool(tool$llmEvalsGetArtifacts);
+    tool(tool$llmEvalsImportAgentRun);
+    tool(tool$llmEvalsAppendAgentTrials);
     tool(tool$llmEvalsImportHistoricalResults);
     tool(tool$llmEvalsCreateSchedule);
     tool(tool$llmEvalsListSchedules);