@meetkai/mka1 0.55.24 → 0.55.26

package/src/mcp-server/tools/billingSetOrganizationBillingEnabled.ts

@@ -14,7 +14,8 @@ export const tool$billingSetOrganizationBillingEnabled: ToolDefinition<
   typeof args
 > = {
   name: "billing-set-organization-billing-enabled",
-  description: ``,
+  description:
+    `Enable or disable organization billing. Requires a human cluster administrator. API keys require both write:billing and the bootstrap-only write:cluster scope. Service-account keys cannot enable or disable billing.`,
   args,
   tool: async (client, args, ctx) => {
     const [result, apiCall] = await billingSetOrganizationBillingEnabled(

package/src/mcp-server/tools/billingUpdateOrganizationBillingProfile.ts

@@ -14,7 +14,8 @@ export const tool$billingUpdateOrganizationBillingProfile: ToolDefinition<
   typeof args
 > = {
   name: "billing-update-organization-billing-profile",
-  description: ``,
+  description:
+    `Update the bound organization's billing profile. API keys require write:billing and either a human owner/admin or delegated service account. The organization is resolved from the API key.`,
   args,
   tool: async (client, args, ctx) => {
     const [result, apiCall] = await billingUpdateOrganizationBillingProfile(

package/src/mcp-server/tools/llmAuditGetRequestTrace.ts

@@ -0,0 +1,37 @@
+/*
+ * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
+ */
+
+import { llmAuditGetRequestTrace } from "../../funcs/llmAuditGetRequestTrace.js";
+import * as operations from "../../models/operations/index.js";
+import { formatResult, ToolDefinition } from "../tools.js";
+
+const args = {
+  request: operations.GetRequestTraceRequest$inboundSchema,
+};
+
+export const tool$llmAuditGetRequestTrace: ToolDefinition<typeof args> = {
+  name: "llm-audit-get-request-trace",
+  description: `Get a request trace
+
+Returns the end-to-end trace of one API request: its timing at the API gateway, every response it created (including agents it called), and each response's model invocations and tool calls with durations. Look it up by the \`X-Request-ID\` returned on every response, or by the API gateway's own request ID. Org owners and admins only, scoped to their own organization. Requests audited before request IDs were recorded show their gateway timing but no responses.`,
+  args,
+  tool: async (client, args, ctx) => {
+    const [result, apiCall] = await llmAuditGetRequestTrace(
+      client,
+      args.request,
+      { fetchOptions: { signal: ctx.mcpReq.signal } },
+    ).$inspect();
+
+    if (!result.ok) {
+      return {
+        content: [{ type: "text", text: result.error.message }],
+        isError: true,
+      };
+    }
+
+    const value = result.value;
+
+    return formatResult(value, apiCall);
+  },
+};

package/src/models/components/agent.ts

@@ -32,6 +32,12 @@ export const AgentObject = {
 } as const;
 export type AgentObject = ClosedEnum<typeof AgentObject>;
 
+export const ConcurrencyOverflow = {
+  Reject: "reject",
+  Queue: "queue",
+} as const;
+export type ConcurrencyOverflow = ClosedEnum<typeof ConcurrencyOverflow>;
+
 export type Agent = {
   object: AgentObject;
   id: string;
@@ -47,6 +53,8 @@ export type Agent = {
   parallelToolCalls: boolean;
   maxToolCalls: number | null;
   fallbackModels: Array<string>;
+  maxConcurrentRuns: number | null;
+  concurrencyOverflow: ConcurrencyOverflow;
   text: AgentTextConfiguration | null;
   reasoning: AgentReasoningConfiguration | null;
   metadata: { [k: string]: string };
@@ -75,6 +83,15 @@ export const AgentObject$inboundSchema: z.ZodNativeEnum<typeof AgentObject> = z
 export const AgentObject$outboundSchema: z.ZodNativeEnum<typeof AgentObject> =
   AgentObject$inboundSchema;
 
+/** @internal */
+export const ConcurrencyOverflow$inboundSchema: z.ZodNativeEnum<
+  typeof ConcurrencyOverflow
+> = z.nativeEnum(ConcurrencyOverflow);
+/** @internal */
+export const ConcurrencyOverflow$outboundSchema: z.ZodNativeEnum<
+  typeof ConcurrencyOverflow
+> = ConcurrencyOverflow$inboundSchema;
+
 /** @internal */
 export const Agent$inboundSchema: z.ZodType<Agent, z.ZodTypeDef, unknown> = z
   .object({
@@ -89,6 +106,8 @@ export const Agent$inboundSchema: z.ZodType<Agent, z.ZodTypeDef, unknown> = z
     parallel_tool_calls: z.boolean(),
     max_tool_calls: z.nullable(z.number().int()),
     fallback_models: z.array(z.string()),
+    max_concurrent_runs: z.nullable(z.number().int()),
+    concurrency_overflow: ConcurrencyOverflow$inboundSchema,
     text: z.nullable(AgentTextConfiguration$inboundSchema),
     reasoning: z.nullable(AgentReasoningConfiguration$inboundSchema),
     metadata: z.record(z.string()),
@@ -113,6 +132,8 @@ export const Agent$inboundSchema: z.ZodType<Agent, z.ZodTypeDef, unknown> = z
       "parallel_tool_calls": "parallelToolCalls",
       "max_tool_calls": "maxToolCalls",
       "fallback_models": "fallbackModels",
+      "max_concurrent_runs": "maxConcurrentRuns",
+      "concurrency_overflow": "concurrencyOverflow",
       "created_at": "createdAt",
       "updated_at": "updatedAt",
       "deleted_at": "deletedAt",
@@ -134,6 +155,8 @@ export type Agent$Outbound = {
   parallel_tool_calls: boolean;
   max_tool_calls: number | null;
   fallback_models: Array<string>;
+  max_concurrent_runs: number | null;
+  concurrency_overflow: string;
   text: AgentTextConfiguration$Outbound | null;
   reasoning: AgentReasoningConfiguration$Outbound | null;
   metadata: { [k: string]: string };
@@ -163,6 +186,8 @@ export const Agent$outboundSchema: z.ZodType<
   parallelToolCalls: z.boolean(),
   maxToolCalls: z.nullable(z.number().int()),
   fallbackModels: z.array(z.string()),
+  maxConcurrentRuns: z.nullable(z.number().int()),
+  concurrencyOverflow: ConcurrencyOverflow$outboundSchema,
   text: z.nullable(AgentTextConfiguration$outboundSchema),
   reasoning: z.nullable(AgentReasoningConfiguration$outboundSchema),
   metadata: z.record(z.string()),
@@ -179,6 +204,8 @@ export const Agent$outboundSchema: z.ZodType<
     parallelToolCalls: "parallel_tool_calls",
     maxToolCalls: "max_tool_calls",
     fallbackModels: "fallback_models",
+    maxConcurrentRuns: "max_concurrent_runs",
+    concurrencyOverflow: "concurrency_overflow",
     createdAt: "created_at",
     updatedAt: "updated_at",
     deletedAt: "deleted_at",

package/src/models/components/agentrun.ts

@@ -15,9 +15,10 @@ export const AgentRunObject = {
 export type AgentRunObject = ClosedEnum<typeof AgentRunObject>;
 
 /**
- * Terminal outcomes stay distinct: cancelled is caller-requested; incomplete means a limit or interruption stopped the run. gateway_response.incomplete_details records the reason, including max_tokens or max_tool_calls. Stored failed rows with one of these gateway outcomes are serialized with the gateway outcome, retaining any connector error and completion timestamp.
+ * waiting means the run is held for one of the agent's max_concurrent_runs slots and has not been sent to the model; it can be cancelled. Terminal outcomes stay distinct: cancelled is caller-requested; incomplete means a limit or interruption stopped the run. gateway_response.incomplete_details records the reason, including max_tokens or max_tool_calls. Stored failed rows with one of these gateway outcomes are serialized with the gateway outcome, retaining any connector error and completion timestamp.
  */
 export const AgentRunStatus = {
+  Waiting: "waiting",
   Queued: "queued",
   Submitting: "submitting",
   Completed: "completed",
@@ -26,7 +27,7 @@ export const AgentRunStatus = {
   Incomplete: "incomplete",
 } as const;
 /**
- * Terminal outcomes stay distinct: cancelled is caller-requested; incomplete means a limit or interruption stopped the run. gateway_response.incomplete_details records the reason, including max_tokens or max_tool_calls. Stored failed rows with one of these gateway outcomes are serialized with the gateway outcome, retaining any connector error and completion timestamp.
+ * waiting means the run is held for one of the agent's max_concurrent_runs slots and has not been sent to the model; it can be cancelled. Terminal outcomes stay distinct: cancelled is caller-requested; incomplete means a limit or interruption stopped the run. gateway_response.incomplete_details records the reason, including max_tokens or max_tool_calls. Stored failed rows with one of these gateway outcomes are serialized with the gateway outcome, retaining any connector error and completion timestamp.
  */
 export type AgentRunStatus = ClosedEnum<typeof AgentRunStatus>;
 
@@ -39,7 +40,7 @@ export type AgentRun = {
    */
   agentVersion: number | null;
   /**
-   * Terminal outcomes stay distinct: cancelled is caller-requested; incomplete means a limit or interruption stopped the run. gateway_response.incomplete_details records the reason, including max_tokens or max_tool_calls. Stored failed rows with one of these gateway outcomes are serialized with the gateway outcome, retaining any connector error and completion timestamp.
+   * waiting means the run is held for one of the agent's max_concurrent_runs slots and has not been sent to the model; it can be cancelled. Terminal outcomes stay distinct: cancelled is caller-requested; incomplete means a limit or interruption stopped the run. gateway_response.incomplete_details records the reason, including max_tokens or max_tool_calls. Stored failed rows with one of these gateway outcomes are serialized with the gateway outcome, retaining any connector error and completion timestamp.
    */
   status: AgentRunStatus;
   /**

package/src/models/components/createagentrequest.ts

@@ -5,6 +5,7 @@
 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";
 import {
@@ -26,6 +27,20 @@ import {
   ToolDefinition$outboundSchema,
 } from "./tooldefinition.js";
 
+/**
+ * What a run beyond max_concurrent_runs gets. reject (the default) refuses it with 429 and error code agent_concurrency_limit. queue stores it as waiting and starts it, oldest first, when a run finishes, with the agent configuration current at that time.
+ */
+export const CreateAgentRequestConcurrencyOverflow = {
+  Reject: "reject",
+  Queue: "queue",
+} as const;
+/**
+ * What a run beyond max_concurrent_runs gets. reject (the default) refuses it with 429 and error code agent_concurrency_limit. queue stores it as waiting and starts it, oldest first, when a run finishes, with the agent configuration current at that time.
+ */
+export type CreateAgentRequestConcurrencyOverflow = ClosedEnum<
+  typeof CreateAgentRequestConcurrencyOverflow
+>;
+
 export type CreateAgentRequest = {
   name: string;
   description?: string | null | undefined;
@@ -42,11 +57,29 @@ export type CreateAgentRequest = {
    * Ordered chain of model IDs the gateway tries when a model call fails on the model before it; forwarded unchanged to the Responses API, which validates it. Empty (the default) means the run keeps the gateway's own fallback rules for the model.
    */
   fallbackModels?: Array<string> | undefined;
+  /**
+   * Most runs of this agent that may execute at once. Null (the default) means no limit. Not part of the versioned configuration: a rollback keeps the current limit, and changing only this field or concurrency_overflow creates no new version.
+   */
+  maxConcurrentRuns?: number | null | undefined;
+  /**
+   * What a run beyond max_concurrent_runs gets. reject (the default) refuses it with 429 and error code agent_concurrency_limit. queue stores it as waiting and starts it, oldest first, when a run finishes, with the agent configuration current at that time.
+   */
+  concurrencyOverflow?: CreateAgentRequestConcurrencyOverflow | undefined;
   text?: AgentTextConfiguration | null | undefined;
   reasoning?: AgentReasoningConfiguration | null | undefined;
   metadata?: { [k: string]: string } | undefined;
 };
 
+/** @internal */
+export const CreateAgentRequestConcurrencyOverflow$inboundSchema:
+  z.ZodNativeEnum<typeof CreateAgentRequestConcurrencyOverflow> = z.nativeEnum(
+    CreateAgentRequestConcurrencyOverflow,
+  );
+/** @internal */
+export const CreateAgentRequestConcurrencyOverflow$outboundSchema:
+  z.ZodNativeEnum<typeof CreateAgentRequestConcurrencyOverflow> =
+    CreateAgentRequestConcurrencyOverflow$inboundSchema;
+
 /** @internal */
 export const CreateAgentRequest$inboundSchema: z.ZodType<
   CreateAgentRequest,
@@ -62,6 +95,9 @@ export const CreateAgentRequest$inboundSchema: z.ZodType<
   parallel_tool_calls: z.boolean().default(true),
   max_tool_calls: z.nullable(z.number().int()).optional(),
   fallback_models: z.array(z.string()).optional(),
+  max_concurrent_runs: z.nullable(z.number().int()).optional(),
+  concurrency_overflow: CreateAgentRequestConcurrencyOverflow$inboundSchema
+    .optional(),
   text: z.nullable(AgentTextConfiguration$inboundSchema).optional(),
   reasoning: z.nullable(AgentReasoningConfiguration$inboundSchema).optional(),
   metadata: z.record(z.string()).optional(),
@@ -71,6 +107,8 @@ export const CreateAgentRequest$inboundSchema: z.ZodType<
     "parallel_tool_calls": "parallelToolCalls",
     "max_tool_calls": "maxToolCalls",
     "fallback_models": "fallbackModels",
+    "max_concurrent_runs": "maxConcurrentRuns",
+    "concurrency_overflow": "concurrencyOverflow",
   });
 });
 /** @internal */
@@ -84,6 +122,8 @@ export type CreateAgentRequest$Outbound = {
   parallel_tool_calls: boolean;
   max_tool_calls?: number | null | undefined;
   fallback_models?: Array<string> | undefined;
+  max_concurrent_runs?: number | null | undefined;
+  concurrency_overflow?: string | undefined;
   text?: AgentTextConfiguration$Outbound | null | undefined;
   reasoning?: AgentReasoningConfiguration$Outbound | null | undefined;
   metadata?: { [k: string]: string } | undefined;
@@ -104,6 +144,9 @@ export const CreateAgentRequest$outboundSchema: z.ZodType<
   parallelToolCalls: z.boolean().default(true),
   maxToolCalls: z.nullable(z.number().int()).optional(),
   fallbackModels: z.array(z.string()).optional(),
+  maxConcurrentRuns: z.nullable(z.number().int()).optional(),
+  concurrencyOverflow: CreateAgentRequestConcurrencyOverflow$outboundSchema
+    .optional(),
   text: z.nullable(AgentTextConfiguration$outboundSchema).optional(),
   reasoning: z.nullable(AgentReasoningConfiguration$outboundSchema).optional(),
   metadata: z.record(z.string()).optional(),
@@ -113,6 +156,8 @@ export const CreateAgentRequest$outboundSchema: z.ZodType<
     parallelToolCalls: "parallel_tool_calls",
     maxToolCalls: "max_tool_calls",
     fallbackModels: "fallback_models",
+    maxConcurrentRuns: "max_concurrent_runs",
+    concurrencyOverflow: "concurrency_overflow",
   });
 });
 

package/src/models/components/createvectorstorerequest.ts

@@ -87,7 +87,7 @@ export type CreateVectorStoreRequest = {
    */
   embeddingModel?: string | undefined;
   /**
-   * The number of dimensions for the embedding vectors. Only supported for models with flexible dimensions. If not specified, uses the model's default dimensions.
+   * The number of dimensions for the embedding vectors. Must equal the model's size when it is fixed, or lie within its range (`embedding_dimension_range` on `GET /models/{model_id}`). If not specified, uses the model's default dimensions. The store stays searchable only while the model's declared dimensions still include this size and the model returns embeddings of this width.
    */
   embeddingDimensions?: number | undefined;
   /**

package/src/models/components/embeddingmodellimits.ts

@@ -13,15 +13,15 @@ import { SDKValidationError } from "../errors/sdkvalidationerror.js";
  */
 export type EmbeddingModelLimits = {
   /**
-   * Maximum number of inputs allowed in a single batch request
+   * Maximum number of inputs allowed in a single batch request: the model's batch size or the gateway-wide 10,000-input limit, whichever is lower, on `POST /embeddings`. Batch `/v1/embeddings` lines are held only to the model's batch size.
    */
   maxBatchSize: number;
   /**
-   * Maximum tokens per individual input (derived from contextWindow)
+   * Maximum tokens per individual input: the model's context window. Null when the registry records no context window: no token cap applies beyond what `max_input_length` implies. For a window long enough that its estimated character length exceeds the gateway-wide character limit, `max_input_length` binds first. Advisory on `/embeddings`, which accepts only strings and enforces `max_input_length`.
    */
-  maxInputTokens: number;
+  maxInputTokens: number | null;
   /**
-   * Maximum characters per individual input (contextWindow * 2)
+   * Maximum characters per individual input: the cap derived from the model's context window or the gateway-wide 100,000-character limit, whichever is lower, on `POST /embeddings`. Batch `/v1/embeddings` lines are held only to the model's cap.
    */
   maxInputLength: number;
 };
@@ -32,9 +32,9 @@ export const EmbeddingModelLimits$inboundSchema: z.ZodType<
   z.ZodTypeDef,
   unknown
 > = z.object({
-  max_batch_size: z.number(),
-  max_input_tokens: z.number(),
-  max_input_length: z.number(),
+  max_batch_size: z.number().int(),
+  max_input_tokens: z.nullable(z.number().int()),
+  max_input_length: z.number().int(),
 }).transform((v) => {
   return remap$(v, {
     "max_batch_size": "maxBatchSize",
@@ -45,7 +45,7 @@ export const EmbeddingModelLimits$inboundSchema: z.ZodType<
 /** @internal */
 export type EmbeddingModelLimits$Outbound = {
   max_batch_size: number;
-  max_input_tokens: number;
+  max_input_tokens: number | null;
   max_input_length: number;
 };
 
@@ -55,9 +55,9 @@ export const EmbeddingModelLimits$outboundSchema: z.ZodType<
   z.ZodTypeDef,
   EmbeddingModelLimits
 > = z.object({
-  maxBatchSize: z.number(),
-  maxInputTokens: z.number(),
-  maxInputLength: z.number(),
+  maxBatchSize: z.number().int(),
+  maxInputTokens: z.nullable(z.number().int()),
+  maxInputLength: z.number().int(),
 }).transform((v) => {
   return remap$(v, {
     maxBatchSize: "max_batch_size",

package/src/models/components/embeddingsrequest.ts

@@ -39,7 +39,7 @@ export type EmbeddingsRequest = {
    */
   model: string;
   /**
-   * The number of dimensions the resulting output embeddings should have. Only supported in certain models.
+   * The number of dimensions the resulting output embeddings should have. Ignored for a model with a fixed size, which always returns that size. For a model with a range it must lie within the range (`embedding_dimension_range` on `GET /models/{model_id}`), or the request fails with 400.
    */
   dimensions?: number | undefined;
   /**

package/src/models/components/updateagentrequest.ts

@@ -5,6 +5,7 @@
 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";
 import {
@@ -26,6 +27,20 @@ import {
   ToolDefinition$outboundSchema,
 } from "./tooldefinition.js";
 
+/**
+ * What a run beyond max_concurrent_runs gets. reject (the default) refuses it with 429 and error code agent_concurrency_limit. queue stores it as waiting and starts it, oldest first, when a run finishes, with the agent configuration current at that time.
+ */
+export const UpdateAgentRequestConcurrencyOverflow = {
+  Reject: "reject",
+  Queue: "queue",
+} as const;
+/**
+ * What a run beyond max_concurrent_runs gets. reject (the default) refuses it with 429 and error code agent_concurrency_limit. queue stores it as waiting and starts it, oldest first, when a run finishes, with the agent configuration current at that time.
+ */
+export type UpdateAgentRequestConcurrencyOverflow = ClosedEnum<
+  typeof UpdateAgentRequestConcurrencyOverflow
+>;
+
 /**
  * Updates only provided fields. Omitted fields preserve their saved values; empty updates are rejected.
  */
@@ -45,11 +60,29 @@ export type UpdateAgentRequest = {
    * Ordered chain of model IDs the gateway tries when a model call fails on the model before it; forwarded unchanged to the Responses API, which validates it. Empty (the default) means the run keeps the gateway's own fallback rules for the model.
    */
   fallbackModels?: Array<string> | undefined;
+  /**
+   * Most runs of this agent that may execute at once. Null (the default) means no limit. Not part of the versioned configuration: a rollback keeps the current limit, and changing only this field or concurrency_overflow creates no new version.
+   */
+  maxConcurrentRuns?: number | null | undefined;
+  /**
+   * What a run beyond max_concurrent_runs gets. reject (the default) refuses it with 429 and error code agent_concurrency_limit. queue stores it as waiting and starts it, oldest first, when a run finishes, with the agent configuration current at that time.
+   */
+  concurrencyOverflow?: UpdateAgentRequestConcurrencyOverflow | undefined;
   text?: AgentTextConfiguration | null | undefined;
   reasoning?: AgentReasoningConfiguration | null | undefined;
   metadata?: { [k: string]: string } | undefined;
 };
 
+/** @internal */
+export const UpdateAgentRequestConcurrencyOverflow$inboundSchema:
+  z.ZodNativeEnum<typeof UpdateAgentRequestConcurrencyOverflow> = z.nativeEnum(
+    UpdateAgentRequestConcurrencyOverflow,
+  );
+/** @internal */
+export const UpdateAgentRequestConcurrencyOverflow$outboundSchema:
+  z.ZodNativeEnum<typeof UpdateAgentRequestConcurrencyOverflow> =
+    UpdateAgentRequestConcurrencyOverflow$inboundSchema;
+
 /** @internal */
 export const UpdateAgentRequest$inboundSchema: z.ZodType<
   UpdateAgentRequest,
@@ -65,6 +98,9 @@ export const UpdateAgentRequest$inboundSchema: z.ZodType<
   parallel_tool_calls: z.boolean().optional(),
   max_tool_calls: z.nullable(z.number().int()).optional(),
   fallback_models: z.array(z.string()).optional(),
+  max_concurrent_runs: z.nullable(z.number().int()).optional(),
+  concurrency_overflow: UpdateAgentRequestConcurrencyOverflow$inboundSchema
+    .optional(),
   text: z.nullable(AgentTextConfiguration$inboundSchema).optional(),
   reasoning: z.nullable(AgentReasoningConfiguration$inboundSchema).optional(),
   metadata: z.record(z.string()).optional(),
@@ -74,6 +110,8 @@ export const UpdateAgentRequest$inboundSchema: z.ZodType<
     "parallel_tool_calls": "parallelToolCalls",
     "max_tool_calls": "maxToolCalls",
     "fallback_models": "fallbackModels",
+    "max_concurrent_runs": "maxConcurrentRuns",
+    "concurrency_overflow": "concurrencyOverflow",
   });
 });
 /** @internal */
@@ -87,6 +125,8 @@ export type UpdateAgentRequest$Outbound = {
   parallel_tool_calls?: boolean | undefined;
   max_tool_calls?: number | null | undefined;
   fallback_models?: Array<string> | undefined;
+  max_concurrent_runs?: number | null | undefined;
+  concurrency_overflow?: string | undefined;
   text?: AgentTextConfiguration$Outbound | null | undefined;
   reasoning?: AgentReasoningConfiguration$Outbound | null | undefined;
   metadata?: { [k: string]: string } | undefined;
@@ -107,6 +147,9 @@ export const UpdateAgentRequest$outboundSchema: z.ZodType<
   parallelToolCalls: z.boolean().optional(),
   maxToolCalls: z.nullable(z.number().int()).optional(),
   fallbackModels: z.array(z.string()).optional(),
+  maxConcurrentRuns: z.nullable(z.number().int()).optional(),
+  concurrencyOverflow: UpdateAgentRequestConcurrencyOverflow$outboundSchema
+    .optional(),
   text: z.nullable(AgentTextConfiguration$outboundSchema).optional(),
   reasoning: z.nullable(AgentReasoningConfiguration$outboundSchema).optional(),
   metadata: z.record(z.string()).optional(),
@@ -116,6 +159,8 @@ export const UpdateAgentRequest$outboundSchema: z.ZodType<
     parallelToolCalls: "parallel_tool_calls",
     maxToolCalls: "max_tool_calls",
     fallbackModels: "fallback_models",
+    maxConcurrentRuns: "max_concurrent_runs",
+    concurrencyOverflow: "concurrency_overflow",
   });
 });
 

package/src/models/components/vectorstore.ts

@@ -88,7 +88,7 @@ export type VectorStore = {
    */
   embeddingModel: string | null;
   /**
-   * The number of dimensions for the embedding vectors in this vector store. Null for legacy vector stores.
+   * The number of dimensions for the embedding vectors in this vector store. Null for legacy vector stores. If the embedding model's declared dimensions stop including this size, or the model returns embeddings of another width, search fails with 412 and new files fail to ingest.
    */
   embeddingDimensions: number | null;
   /**

package/src/models/operations/getmodel.ts

@@ -160,7 +160,7 @@ export type GetModelResponseBody = {
    */
   outputModalities: Array<string>;
   /**
-   * Context window in tokens, mirroring the registry's `contextWindow`. For an LLM it is the input plus output tokens one request can span: the Responses API trims the oldest conversation items to fit it, reserving nothing for the output, unless the request sets `truncation` to `disabled`, which rejects an oversized request instead; a separate request-size limit in bytes applies the same way. Chat completions forwards requests unchanged. For an embeddings model `/embeddings` derives its per-input character cap from it, which `GET /embeddings/models` reports as `max_input_length`; a gateway-wide per-input ceiling also applies. On other model types the gateway does not read it, so a recorded value is informational only and not enforced. Null when the registry does not record one.
+   * Context window in tokens, mirroring the registry's `contextWindow`. For an LLM it is the input plus output tokens one request can span: the Responses API trims the oldest conversation items to fit it, reserving nothing for the output, unless the request sets `truncation` to `disabled`, which rejects an oversized request instead; a separate request-size limit in bytes applies the same way. Chat completions forwards requests unchanged. For an embeddings model `/embeddings` derives a per-input character cap from it, alongside a gateway-wide per-input ceiling; `GET /embeddings/models` reports whichever binds first as `max_input_length`. On other model types the gateway does not read it, so a recorded value is informational only and not enforced. Null when the registry does not record one.
    */
   contextWindow: number | null;
   /**
@@ -172,7 +172,7 @@ export type GetModelResponseBody = {
    */
   minImageResolution: MinImageResolution | null;
   /**
-   * Output vector sizes declared for this model, mirroring the registry's `embeddingLimits.dimensions`, which is normally recorded only on embeddings models. A model with a single fixed size reports it as `min`, `max` and `default` alike, so when `min` equals `max` the size is fixed and a request should omit `dimensions`. Vector stores reject a size outside the range. `/embeddings` does not check it: it drops a requested `dimensions` for a fixed-size model and otherwise forwards it to the upstream. Null when the registry does not record dimensions; a vector store using such a model must then be created with an explicit size (`embedding_dimensions` on `POST /vector_stores`). Distinct from the vector store's own `embedding_dimensions`, which is a single size.
+   * Output vector sizes declared for this model, mirroring the registry's `embeddingLimits.dimensions`, which is normally recorded only on embeddings models. When `min` equals `max` the size is fixed: `/embeddings` ignores a requested `dimensions` and returns that size, and vector stores accept only that size. Otherwise a requested size must lie within the range: `/embeddings` and vector stores reject one outside it with 400. Null when the registry does not record dimensions; a vector store using such a model must then be created with an explicit size (`embedding_dimensions` on `POST /vector_stores`). Distinct from the vector store's own `embedding_dimensions`, which is a single size.
    */
   embeddingDimensionRange: EmbeddingDimensionRange | null;
   /**