@meetkai/mka1 0.51.2 → 0.51.4

package/dist/esm/models/operations/updatescaling.d.ts

@@ -8,7 +8,7 @@ export type UpdateScalingRequest = {
      * Optional external end-user identifier forwarded by the API gateway.
      */
     xOnBehalfOf?: string | undefined;
-    scaling: components.Scaling;
+    deploymentScalingUpdate: components.DeploymentScalingUpdate;
 };
 /** @internal */
 export declare const UpdateScalingRequest$inboundSchema: z.ZodType<UpdateScalingRequest, z.ZodTypeDef, unknown>;
@@ -16,7 +16,7 @@ export declare const UpdateScalingRequest$inboundSchema: z.ZodType<UpdateScaling
 export type UpdateScalingRequest$Outbound = {
     deployment_id: string;
     "X-On-Behalf-Of"?: string | undefined;
-    Scaling: components.Scaling$Outbound;
+    DeploymentScalingUpdate: components.DeploymentScalingUpdate$Outbound;
 };
 /** @internal */
 export declare const UpdateScalingRequest$outboundSchema: z.ZodType<UpdateScalingRequest$Outbound, z.ZodTypeDef, UpdateScalingRequest>;

package/dist/esm/models/operations/updatescaling.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"updatescaling.d.ts","sourceRoot":"","sources":["../../../../src/models/operations/updatescaling.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,QAAQ,CAAC;AAG5B,OAAO,EAAE,MAAM,IAAI,eAAe,EAAE,MAAM,mBAAmB,CAAC;AAC9D,OAAO,KAAK,UAAU,MAAM,wBAAwB,CAAC;AACrD,OAAO,EAAE,kBAAkB,EAAE,MAAM,iCAAiC,CAAC;AAErE,MAAM,MAAM,oBAAoB,GAAG;IACjC,YAAY,EAAE,MAAM,CAAC;IACrB;;OAEG;IACH,WAAW,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IACjC,OAAO,EAAE,UAAU,CAAC,OAAO,CAAC;CAC7B,CAAC;AAEF,gBAAgB;AAChB,eAAO,MAAM,kCAAkC,EAAE,CAAC,CAAC,OAAO,CACxD,oBAAoB,EACpB,CAAC,CAAC,UAAU,EACZ,OAAO,CAWP,CAAC;AACH,gBAAgB;AAChB,MAAM,MAAM,6BAA6B,GAAG;IAC1C,aAAa,EAAE,MAAM,CAAC;IACtB,gBAAgB,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IACtC,OAAO,EAAE,UAAU,CAAC,gBAAgB,CAAC;CACtC,CAAC;AAEF,gBAAgB;AAChB,eAAO,MAAM,mCAAmC,EAAE,CAAC,CAAC,OAAO,CACzD,6BAA6B,EAC7B,CAAC,CAAC,UAAU,EACZ,oBAAoB,CAWpB,CAAC;AAEH,wBAAgB,0BAA0B,CACxC,oBAAoB,EAAE,oBAAoB,GACzC,MAAM,CAIR;AACD,wBAAgB,4BAA4B,CAC1C,UAAU,EAAE,MAAM,GACjB,eAAe,CAAC,oBAAoB,EAAE,kBAAkB,CAAC,CAM3D"}
+{"version":3,"file":"updatescaling.d.ts","sourceRoot":"","sources":["../../../../src/models/operations/updatescaling.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,QAAQ,CAAC;AAG5B,OAAO,EAAE,MAAM,IAAI,eAAe,EAAE,MAAM,mBAAmB,CAAC;AAC9D,OAAO,KAAK,UAAU,MAAM,wBAAwB,CAAC;AACrD,OAAO,EAAE,kBAAkB,EAAE,MAAM,iCAAiC,CAAC;AAErE,MAAM,MAAM,oBAAoB,GAAG;IACjC,YAAY,EAAE,MAAM,CAAC;IACrB;;OAEG;IACH,WAAW,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IACjC,uBAAuB,EAAE,UAAU,CAAC,uBAAuB,CAAC;CAC7D,CAAC;AAEF,gBAAgB;AAChB,eAAO,MAAM,kCAAkC,EAAE,CAAC,CAAC,OAAO,CACxD,oBAAoB,EACpB,CAAC,CAAC,UAAU,EACZ,OAAO,CAWP,CAAC;AACH,gBAAgB;AAChB,MAAM,MAAM,6BAA6B,GAAG;IAC1C,aAAa,EAAE,MAAM,CAAC;IACtB,gBAAgB,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IACtC,uBAAuB,EAAE,UAAU,CAAC,gCAAgC,CAAC;CACtE,CAAC;AAEF,gBAAgB;AAChB,eAAO,MAAM,mCAAmC,EAAE,CAAC,CAAC,OAAO,CACzD,6BAA6B,EAC7B,CAAC,CAAC,UAAU,EACZ,oBAAoB,CAWpB,CAAC;AAEH,wBAAgB,0BAA0B,CACxC,oBAAoB,EAAE,oBAAoB,GACzC,MAAM,CAIR;AACD,wBAAgB,4BAA4B,CAC1C,UAAU,EAAE,MAAM,GACjB,eAAe,CAAC,oBAAoB,EAAE,kBAAkB,CAAC,CAM3D"}

package/dist/esm/models/operations/updatescaling.js

@@ -9,24 +9,24 @@ import * as components from "../components/index.js";
 export const UpdateScalingRequest$inboundSchema = z.object({
     deployment_id: z.string(),
     "X-On-Behalf-Of": z.string().optional(),
-    Scaling: components.Scaling$inboundSchema,
+    DeploymentScalingUpdate: components.DeploymentScalingUpdate$inboundSchema,
 }).transform((v) => {
     return remap$(v, {
         "deployment_id": "deploymentId",
         "X-On-Behalf-Of": "xOnBehalfOf",
-        "Scaling": "scaling",
+        "DeploymentScalingUpdate": "deploymentScalingUpdate",
     });
 });
 /** @internal */
 export const UpdateScalingRequest$outboundSchema = z.object({
     deploymentId: z.string(),
     xOnBehalfOf: z.string().optional(),
-    scaling: components.Scaling$outboundSchema,
+    deploymentScalingUpdate: components.DeploymentScalingUpdate$outboundSchema,
 }).transform((v) => {
     return remap$(v, {
         deploymentId: "deployment_id",
         xOnBehalfOf: "X-On-Behalf-Of",
-        scaling: "Scaling",
+        deploymentScalingUpdate: "DeploymentScalingUpdate",
     });
 });
 export function updateScalingRequestToJSON(updateScalingRequest) {

package/dist/esm/models/operations/updatescaling.js.map

@@ -1 +1 @@
-{"version":3,"file":"updatescaling.js","sourceRoot":"","sources":["../../../../src/models/operations/updatescaling.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,QAAQ,CAAC;AAC5B,OAAO,EAAE,KAAK,IAAI,MAAM,EAAE,MAAM,yBAAyB,CAAC;AAC1D,OAAO,EAAE,SAAS,EAAE,MAAM,sBAAsB,CAAC;AAEjD,OAAO,KAAK,UAAU,MAAM,wBAAwB,CAAC;AAYrD,gBAAgB;AAChB,MAAM,CAAC,MAAM,kCAAkC,GAI3C,CAAC,CAAC,MAAM,CAAC;IACX,aAAa,EAAE,CAAC,CAAC,MAAM,EAAE;IACzB,gBAAgB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IACvC,OAAO,EAAE,UAAU,CAAC,qBAAqB;CAC1C,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,EAAE;IACjB,OAAO,MAAM,CAAC,CAAC,EAAE;QACf,eAAe,EAAE,cAAc;QAC/B,gBAAgB,EAAE,aAAa;QAC/B,SAAS,EAAE,SAAS;KACrB,CAAC,CAAC;AACL,CAAC,CAAC,CAAC;AAQH,gBAAgB;AAChB,MAAM,CAAC,MAAM,mCAAmC,GAI5C,CAAC,CAAC,MAAM,CAAC;IACX,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE;IACxB,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAClC,OAAO,EAAE,UAAU,CAAC,sBAAsB;CAC3C,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,EAAE;IACjB,OAAO,MAAM,CAAC,CAAC,EAAE;QACf,YAAY,EAAE,eAAe;QAC7B,WAAW,EAAE,gBAAgB;QAC7B,OAAO,EAAE,SAAS;KACnB,CAAC,CAAC;AACL,CAAC,CAAC,CAAC;AAEH,MAAM,UAAU,0BAA0B,CACxC,oBAA0C;IAE1C,OAAO,IAAI,CAAC,SAAS,CACnB,mCAAmC,CAAC,KAAK,CAAC,oBAAoB,CAAC,CAChE,CAAC;AACJ,CAAC;AACD,MAAM,UAAU,4BAA4B,CAC1C,UAAkB;IAElB,OAAO,SAAS,CACd,UAAU,EACV,CAAC,CAAC,EAAE,EAAE,CAAC,kCAAkC,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,EAC9D,kDAAkD,CACnD,CAAC;AACJ,CAAC"}
+{"version":3,"file":"updatescaling.js","sourceRoot":"","sources":["../../../../src/models/operations/updatescaling.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,QAAQ,CAAC;AAC5B,OAAO,EAAE,KAAK,IAAI,MAAM,EAAE,MAAM,yBAAyB,CAAC;AAC1D,OAAO,EAAE,SAAS,EAAE,MAAM,sBAAsB,CAAC;AAEjD,OAAO,KAAK,UAAU,MAAM,wBAAwB,CAAC;AAYrD,gBAAgB;AAChB,MAAM,CAAC,MAAM,kCAAkC,GAI3C,CAAC,CAAC,MAAM,CAAC;IACX,aAAa,EAAE,CAAC,CAAC,MAAM,EAAE;IACzB,gBAAgB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IACvC,uBAAuB,EAAE,UAAU,CAAC,qCAAqC;CAC1E,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,EAAE;IACjB,OAAO,MAAM,CAAC,CAAC,EAAE;QACf,eAAe,EAAE,cAAc;QAC/B,gBAAgB,EAAE,aAAa;QAC/B,yBAAyB,EAAE,yBAAyB;KACrD,CAAC,CAAC;AACL,CAAC,CAAC,CAAC;AAQH,gBAAgB;AAChB,MAAM,CAAC,MAAM,mCAAmC,GAI5C,CAAC,CAAC,MAAM,CAAC;IACX,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE;IACxB,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAClC,uBAAuB,EAAE,UAAU,CAAC,sCAAsC;CAC3E,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,EAAE;IACjB,OAAO,MAAM,CAAC,CAAC,EAAE;QACf,YAAY,EAAE,eAAe;QAC7B,WAAW,EAAE,gBAAgB;QAC7B,uBAAuB,EAAE,yBAAyB;KACnD,CAAC,CAAC;AACL,CAAC,CAAC,CAAC;AAEH,MAAM,UAAU,0BAA0B,CACxC,oBAA0C;IAE1C,OAAO,IAAI,CAAC,SAAS,CACnB,mCAAmC,CAAC,KAAK,CAAC,oBAAoB,CAAC,CAChE,CAAC;AACJ,CAAC;AACD,MAAM,UAAU,4BAA4B,CAC1C,UAAkB;IAElB,OAAO,SAAS,CACd,UAAU,EACV,CAAC,CAAC,EAAE,EAAE,CAAC,kCAAkC,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,EAC9D,kDAAkD,CACnD,CAAC;AACJ,CAAC"}

package/jsr.json

@@ -2,7 +2,7 @@
 
 {
   "name": "@meetkai/mka1",
-  "version": "0.51.2",
+  "version": "0.51.4",
   "exports": {
     ".": "./src/index.ts",    
     "./models/errors": "./src/models/errors/index.ts",    

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@meetkai/mka1",
-  "version": "0.51.2",
+  "version": "0.51.4",
   "author": "Daniel Wilder <dauncy@gmail.com>",
   "homepage": "https://docs.mka1.com",
   "publishConfig": {

package/src/funcs/servingDeploymentsCreate.ts

@@ -178,9 +178,10 @@ async function $do(
     | SDKValidationError
   >(
     M.json(201, components.Deployment$inboundSchema),
-    M.jsonErr(403, errors.ServingError$inboundSchema),
+    M.jsonErr([401, 403, 404, 409], errors.ServingError$inboundSchema),
     M.jsonErr(422, errors.ServingHTTPValidationError$inboundSchema),
-    M.fail([401, "4XX"]),
+    M.jsonErr([500, 503], errors.ServingError$inboundSchema),
+    M.fail("4XX"),
     M.fail("5XX"),
   )(response, req, { extraFields: responseFields });
   if (!result.ok) {

package/src/funcs/servingDeploymentsDelete.ts

@@ -185,9 +185,10 @@ async function $do(
     | SDKValidationError
   >(
     M.nil(204, z.void()),
-    M.jsonErr([403, 404], errors.ServingError$inboundSchema),
+    M.jsonErr([401, 403, 404], errors.ServingError$inboundSchema),
     M.jsonErr(422, errors.ServingHTTPValidationError$inboundSchema),
-    M.fail([401, "4XX"]),
+    M.jsonErr([500, 503], errors.ServingError$inboundSchema),
+    M.fail("4XX"),
     M.fail("5XX"),
   )(response, req, { extraFields: responseFields });
   if (!result.ok) {

package/src/funcs/servingDeploymentsGet.ts

@@ -185,9 +185,10 @@ async function $do(
     | SDKValidationError
   >(
     M.json(200, components.Deployment$inboundSchema),
-    M.jsonErr([403, 404], errors.ServingError$inboundSchema),
+    M.jsonErr([401, 403, 404], errors.ServingError$inboundSchema),
     M.jsonErr(422, errors.ServingHTTPValidationError$inboundSchema),
-    M.fail([401, "4XX"]),
+    M.jsonErr([500, 503], errors.ServingError$inboundSchema),
+    M.fail("4XX"),
     M.fail("5XX"),
   )(response, req, { extraFields: responseFields });
   if (!result.ok) {

package/src/funcs/servingDeploymentsList.ts

@@ -183,9 +183,10 @@ async function $do(
     | SDKValidationError
   >(
     M.json(200, components.PageDeployment$inboundSchema),
-    M.jsonErr(403, errors.ServingError$inboundSchema),
+    M.jsonErr([401, 403], errors.ServingError$inboundSchema),
     M.jsonErr(422, errors.ServingHTTPValidationError$inboundSchema),
-    M.fail([401, "4XX"]),
+    M.jsonErr([500, 503], errors.ServingError$inboundSchema),
+    M.fail("4XX"),
     M.fail("5XX"),
   )(response, req, { extraFields: responseFields });
   if (!result.ok) {

package/src/funcs/servingDeploymentsListReplicas.ts

@@ -187,6 +187,7 @@ async function $do(
     M.json(200, components.PageReplica$inboundSchema),
     M.jsonErr([403, 404], errors.ServingError$inboundSchema),
     M.jsonErr(422, errors.ServingHTTPValidationError$inboundSchema),
+    M.jsonErr([500, 503], errors.ServingError$inboundSchema),
     M.fail([401, "4XX"]),
     M.fail("5XX"),
   )(response, req, { extraFields: responseFields });

package/src/funcs/servingDeploymentsListRevisions.ts

@@ -185,9 +185,10 @@ async function $do(
     | SDKValidationError
   >(
     M.json(200, components.PageRevision$inboundSchema),
-    M.jsonErr([403, 404], errors.ServingError$inboundSchema),
+    M.jsonErr([401, 403, 404], errors.ServingError$inboundSchema),
     M.jsonErr(422, errors.ServingHTTPValidationError$inboundSchema),
-    M.fail([401, "4XX"]),
+    M.jsonErr([500, 503], errors.ServingError$inboundSchema),
+    M.fail("4XX"),
     M.fail("5XX"),
   )(response, req, { extraFields: responseFields });
   if (!result.ok) {

package/src/funcs/servingDeploymentsUpdate.ts

@@ -186,9 +186,10 @@ async function $do(
     | SDKValidationError
   >(
     M.json(200, components.Deployment$inboundSchema),
-    M.jsonErr([403, 404], errors.ServingError$inboundSchema),
+    M.jsonErr([401, 403, 404], errors.ServingError$inboundSchema),
     M.jsonErr(422, errors.ServingHTTPValidationError$inboundSchema),
-    M.fail([401, "4XX"]),
+    M.jsonErr([500, 503], errors.ServingError$inboundSchema),
+    M.fail("4XX"),
     M.fail("5XX"),
   )(response, req, { extraFields: responseFields });
   if (!result.ok) {

package/src/funcs/servingDeploymentsUpdateScaling.ts

@@ -90,7 +90,9 @@ async function $do(
     return [parsed, { status: "invalid" }];
   }
   const payload = parsed.value;
-  const body = encodeJSON("body", payload.Scaling, { explode: true });
+  const body = encodeJSON("body", payload.DeploymentScalingUpdate, {
+    explode: true,
+  });
 
   const pathParams = {
     deployment_id: encodeSimple("deployment_id", payload.deployment_id, {
@@ -188,6 +190,7 @@ async function $do(
     M.json(200, components.Deployment$inboundSchema),
     M.jsonErr([403, 404], errors.ServingError$inboundSchema),
     M.jsonErr(422, errors.ServingHTTPValidationError$inboundSchema),
+    M.jsonErr([500, 503], errors.ServingError$inboundSchema),
     M.fail([401, "4XX"]),
     M.fail("5XX"),
   )(response, req, { extraFields: responseFields });

package/src/funcs/servingModelsDelete.ts

@@ -183,7 +183,7 @@ async function $do(
     | SDKValidationError
   >(
     M.nil(204, z.void()),
-    M.jsonErr([403, 404], errors.ServingError$inboundSchema),
+    M.jsonErr([403, 404, 409], errors.ServingError$inboundSchema),
     M.jsonErr(422, errors.ServingHTTPValidationError$inboundSchema),
     M.fail([401, "4XX"]),
     M.fail("5XX"),

package/src/funcs/servingSecretsDelete.ts

@@ -183,7 +183,7 @@ async function $do(
     | SDKValidationError
   >(
     M.nil(204, z.void()),
-    M.jsonErr([403, 404], errors.ServingError$inboundSchema),
+    M.jsonErr([403, 404, 409], errors.ServingError$inboundSchema),
     M.jsonErr(422, errors.ServingHTTPValidationError$inboundSchema),
     M.fail([401, "4XX"]),
     M.fail("5XX"),

package/src/lib/config.ts

@@ -65,7 +65,7 @@ export function serverURLFromOptions(options: SDKOptions): URL | null {
 export const SDK_METADATA = {
   language: "typescript",
   openapiDocVersion: "1.1.0",
-  sdkVersion: "0.51.2",
+  sdkVersion: "0.51.4",
   genVersion: "2.918.4",
-  userAgent: "speakeasy-sdk/typescript 0.51.2 2.918.4 1.1.0 @meetkai/mka1",
+  userAgent: "speakeasy-sdk/typescript 0.51.4 2.918.4 1.1.0 @meetkai/mka1",
 } as const;

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

@@ -19,7 +19,7 @@ const routes = buildRouteMap({
 export const app = buildApplication(routes, {
   name: "mcp",
   versionInfo: {
-    currentVersion: "0.51.2",
+    currentVersion: "0.51.4",
   },
 });
 

package/src/mcp-server/server.ts

@@ -350,7 +350,7 @@ export function createMCPServer(deps: {
 }) {
   const server = new McpServer({
     name: "SDK",
-    version: "0.51.2",
+    version: "0.51.4",
   });
 
   const client = new SDKCore({

package/src/models/components/deploymentapiformat.ts

@@ -0,0 +1,31 @@
+/*
+ * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
+ */
+
+import * as z from "zod/v3";
+import { ClosedEnum } from "../../types/enums.js";
+
+/**
+ * Gateway inference surface exposed by the deployment runtime.
+ */
+export const DeploymentAPIFormat = {
+  Responses: "responses",
+  Completions: "completions",
+  Embeddings: "embeddings",
+  Images: "images",
+  Transcriptions: "transcriptions",
+  Tts: "tts",
+} as const;
+/**
+ * Gateway inference surface exposed by the deployment runtime.
+ */
+export type DeploymentAPIFormat = ClosedEnum<typeof DeploymentAPIFormat>;
+
+/** @internal */
+export const DeploymentAPIFormat$inboundSchema: z.ZodNativeEnum<
+  typeof DeploymentAPIFormat
+> = z.nativeEnum(DeploymentAPIFormat);
+/** @internal */
+export const DeploymentAPIFormat$outboundSchema: z.ZodNativeEnum<
+  typeof DeploymentAPIFormat
+> = DeploymentAPIFormat$inboundSchema;

package/src/models/components/deploymentcreate.ts

@@ -13,6 +13,17 @@ import {
   AcceleratorSpec$Outbound,
   AcceleratorSpec$outboundSchema,
 } from "./acceleratorspec.js";
+import {
+  DeploymentAPIFormat,
+  DeploymentAPIFormat$inboundSchema,
+  DeploymentAPIFormat$outboundSchema,
+} from "./deploymentapiformat.js";
+import {
+  DeploymentScalingCreate,
+  DeploymentScalingCreate$inboundSchema,
+  DeploymentScalingCreate$Outbound,
+  DeploymentScalingCreate$outboundSchema,
+} from "./deploymentscalingcreate.js";
 import {
   DeployStrategy,
   DeployStrategy$inboundSchema,
@@ -23,12 +34,6 @@ import {
   Engine$inboundSchema,
   Engine$outboundSchema,
 } from "./engine.js";
-import {
-  Scaling,
-  Scaling$inboundSchema,
-  Scaling$Outbound,
-  Scaling$outboundSchema,
-} from "./scaling.js";
 
 /**
  * Spin up an inference server.
@@ -42,6 +47,10 @@ export type DeploymentCreate = {
    * Request an accelerator by type and count, with optional fallback types.
    */
   accelerator: AcceleratorSpec;
+  /**
+   * Gateway inference surface exposed by the deployment runtime.
+   */
+  apiFormat?: DeploymentAPIFormat | undefined;
   /**
    * Require an endpoint key (separate from the model's own API key)
    */
@@ -55,9 +64,13 @@ export type DeploymentCreate = {
    */
   engineArgs?: Array<string> | undefined;
   /**
-   * Name/ID of a built image; defaults to the engine's stock image
+   * Digest-pinned OCI image for pure BYO mode; absent or null selects managed mode
    */
   image?: string | null | undefined;
+  /**
+   * Hard aggregate hourly spend ceiling for all replicas and rollout overlap
+   */
+  maxUsdHr: number;
   /**
    * ID of a registered model (see Serving Models)
    */
@@ -67,9 +80,17 @@ export type DeploymentCreate = {
    */
   name: string;
   /**
-   * Autoscaling configuration for a deployment.
+   * Container serving port required for pure BYO mode
+   */
+  port?: number | undefined;
+  /**
+   * Managed runtime recipe key; omitted to resolve uniquely from model metadata
+   */
+  runtime?: string | undefined;
+  /**
+   * Create-time scaling. Omitted min_containers defaults to one; explicit zero is preserved.
    */
-  scaling?: Scaling | undefined;
+  scaling?: DeploymentScalingCreate | undefined;
   secrets?: Array<string> | undefined;
   /**
    * How a config change or rollback is rolled out.
@@ -88,32 +109,42 @@ export const DeploymentCreate$inboundSchema: z.ZodType<
   unknown
 > = z.object({
   accelerator: AcceleratorSpec$inboundSchema,
+  api_format: DeploymentAPIFormat$inboundSchema.optional(),
   endpoint_auth: z.boolean().default(true),
   engine: Engine$inboundSchema.optional(),
   engine_args: z.array(z.string()).optional(),
   image: z.nullable(z.string()).optional(),
+  max_usd_hr: z.number(),
   model: z.string(),
   name: z.string(),
-  scaling: Scaling$inboundSchema.optional(),
+  port: z.number().int().optional(),
+  runtime: z.string().optional(),
+  scaling: DeploymentScalingCreate$inboundSchema.optional(),
   secrets: z.array(z.string()).optional(),
   strategy: DeployStrategy$inboundSchema.optional(),
   volume: z.nullable(z.string()).optional(),
 }).transform((v) => {
   return remap$(v, {
+    "api_format": "apiFormat",
     "endpoint_auth": "endpointAuth",
     "engine_args": "engineArgs",
+    "max_usd_hr": "maxUsdHr",
   });
 });
 /** @internal */
 export type DeploymentCreate$Outbound = {
   accelerator: AcceleratorSpec$Outbound;
+  api_format?: string | undefined;
   endpoint_auth: boolean;
   engine?: string | undefined;
   engine_args?: Array<string> | undefined;
   image?: string | null | undefined;
+  max_usd_hr: number;
   model: string;
   name: string;
-  scaling?: Scaling$Outbound | undefined;
+  port?: number | undefined;
+  runtime?: string | undefined;
+  scaling?: DeploymentScalingCreate$Outbound | undefined;
   secrets?: Array<string> | undefined;
   strategy?: string | undefined;
   volume?: string | null | undefined;
@@ -126,20 +157,26 @@ export const DeploymentCreate$outboundSchema: z.ZodType<
   DeploymentCreate
 > = z.object({
   accelerator: AcceleratorSpec$outboundSchema,
+  apiFormat: DeploymentAPIFormat$outboundSchema.optional(),
   endpointAuth: z.boolean().default(true),
   engine: Engine$outboundSchema.optional(),
   engineArgs: z.array(z.string()).optional(),
   image: z.nullable(z.string()).optional(),
+  maxUsdHr: z.number(),
   model: z.string(),
   name: z.string(),
-  scaling: Scaling$outboundSchema.optional(),
+  port: z.number().int().optional(),
+  runtime: z.string().optional(),
+  scaling: DeploymentScalingCreate$outboundSchema.optional(),
   secrets: z.array(z.string()).optional(),
   strategy: DeployStrategy$outboundSchema.optional(),
   volume: z.nullable(z.string()).optional(),
 }).transform((v) => {
   return remap$(v, {
+    apiFormat: "api_format",
     endpointAuth: "endpoint_auth",
     engineArgs: "engine_args",
+    maxUsdHr: "max_usd_hr",
   });
 });
 

package/src/models/components/deploymentscalingcreate.ts

@@ -0,0 +1,90 @@
+/*
+ * 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 { Result as SafeParseResult } from "../../types/fp.js";
+import { SDKValidationError } from "../errors/sdkvalidationerror.js";
+
+/**
+ * Create-time scaling. Omitted min_containers defaults to one; explicit zero is preserved.
+ */
+export type DeploymentScalingCreate = {
+  bufferContainers?: number | undefined;
+  maxConcurrentInputs?: number | undefined;
+  maxContainers?: number | undefined;
+  /**
+   * Desired warm replicas; omitted defaults to one and explicit zero creates a stopped deployment
+   */
+  minContainers?: number | undefined;
+  scaledownWindowS?: number | undefined;
+};
+
+/** @internal */
+export const DeploymentScalingCreate$inboundSchema: z.ZodType<
+  DeploymentScalingCreate,
+  z.ZodTypeDef,
+  unknown
+> = z.object({
+  buffer_containers: z.number().int().default(0),
+  max_concurrent_inputs: z.number().int().default(100),
+  max_containers: z.number().int().default(1),
+  min_containers: z.number().int().default(1),
+  scaledown_window_s: z.number().int().default(300),
+}).transform((v) => {
+  return remap$(v, {
+    "buffer_containers": "bufferContainers",
+    "max_concurrent_inputs": "maxConcurrentInputs",
+    "max_containers": "maxContainers",
+    "min_containers": "minContainers",
+    "scaledown_window_s": "scaledownWindowS",
+  });
+});
+/** @internal */
+export type DeploymentScalingCreate$Outbound = {
+  buffer_containers: number;
+  max_concurrent_inputs: number;
+  max_containers: number;
+  min_containers: number;
+  scaledown_window_s: number;
+};
+
+/** @internal */
+export const DeploymentScalingCreate$outboundSchema: z.ZodType<
+  DeploymentScalingCreate$Outbound,
+  z.ZodTypeDef,
+  DeploymentScalingCreate
+> = z.object({
+  bufferContainers: z.number().int().default(0),
+  maxConcurrentInputs: z.number().int().default(100),
+  maxContainers: z.number().int().default(1),
+  minContainers: z.number().int().default(1),
+  scaledownWindowS: z.number().int().default(300),
+}).transform((v) => {
+  return remap$(v, {
+    bufferContainers: "buffer_containers",
+    maxConcurrentInputs: "max_concurrent_inputs",
+    maxContainers: "max_containers",
+    minContainers: "min_containers",
+    scaledownWindowS: "scaledown_window_s",
+  });
+});
+
+export function deploymentScalingCreateToJSON(
+  deploymentScalingCreate: DeploymentScalingCreate,
+): string {
+  return JSON.stringify(
+    DeploymentScalingCreate$outboundSchema.parse(deploymentScalingCreate),
+  );
+}
+export function deploymentScalingCreateFromJSON(
+  jsonString: string,
+): SafeParseResult<DeploymentScalingCreate, SDKValidationError> {
+  return safeParse(
+    jsonString,
+    (x) => DeploymentScalingCreate$inboundSchema.parse(JSON.parse(x)),
+    `Failed to parse 'DeploymentScalingCreate' from JSON`,
+  );
+}

package/src/models/components/deploymentscalingupdate.ts

@@ -0,0 +1,87 @@
+/*
+ * 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 { Result as SafeParseResult } from "../../types/fp.js";
+import { SDKValidationError } from "../errors/sdkvalidationerror.js";
+
+/**
+ * Sparse scaling patch; omitted fields retain their effective values.
+ */
+export type DeploymentScalingUpdate = {
+  bufferContainers?: number | undefined;
+  maxConcurrentInputs?: number | undefined;
+  maxContainers?: number | undefined;
+  minContainers?: number | undefined;
+  scaledownWindowS?: number | undefined;
+};
+
+/** @internal */
+export const DeploymentScalingUpdate$inboundSchema: z.ZodType<
+  DeploymentScalingUpdate,
+  z.ZodTypeDef,
+  unknown
+> = z.object({
+  buffer_containers: z.number().int().optional(),
+  max_concurrent_inputs: z.number().int().optional(),
+  max_containers: z.number().int().optional(),
+  min_containers: z.number().int().optional(),
+  scaledown_window_s: z.number().int().optional(),
+}).transform((v) => {
+  return remap$(v, {
+    "buffer_containers": "bufferContainers",
+    "max_concurrent_inputs": "maxConcurrentInputs",
+    "max_containers": "maxContainers",
+    "min_containers": "minContainers",
+    "scaledown_window_s": "scaledownWindowS",
+  });
+});
+/** @internal */
+export type DeploymentScalingUpdate$Outbound = {
+  buffer_containers?: number | undefined;
+  max_concurrent_inputs?: number | undefined;
+  max_containers?: number | undefined;
+  min_containers?: number | undefined;
+  scaledown_window_s?: number | undefined;
+};
+
+/** @internal */
+export const DeploymentScalingUpdate$outboundSchema: z.ZodType<
+  DeploymentScalingUpdate$Outbound,
+  z.ZodTypeDef,
+  DeploymentScalingUpdate
+> = z.object({
+  bufferContainers: z.number().int().optional(),
+  maxConcurrentInputs: z.number().int().optional(),
+  maxContainers: z.number().int().optional(),
+  minContainers: z.number().int().optional(),
+  scaledownWindowS: z.number().int().optional(),
+}).transform((v) => {
+  return remap$(v, {
+    bufferContainers: "buffer_containers",
+    maxConcurrentInputs: "max_concurrent_inputs",
+    maxContainers: "max_containers",
+    minContainers: "min_containers",
+    scaledownWindowS: "scaledown_window_s",
+  });
+});
+
+export function deploymentScalingUpdateToJSON(
+  deploymentScalingUpdate: DeploymentScalingUpdate,
+): string {
+  return JSON.stringify(
+    DeploymentScalingUpdate$outboundSchema.parse(deploymentScalingUpdate),
+  );
+}
+export function deploymentScalingUpdateFromJSON(
+  jsonString: string,
+): SafeParseResult<DeploymentScalingUpdate, SDKValidationError> {
+  return safeParse(
+    jsonString,
+    (x) => DeploymentScalingUpdate$inboundSchema.parse(JSON.parse(x)),
+    `Failed to parse 'DeploymentScalingUpdate' from JSON`,
+  );
+}