@kaana.ai/contracts 8.10.0 → 8.12.0

package/dist/index.d.ts

@@ -26,6 +26,7 @@ export * from './public/money.js';
 export * from './public/policy.js';
 export * from './public/providerPreferences.js';
 export * from './public/realtime.js';
+export * from './public/rerank.js';
 export * from './public/responses.js';
 export * from './public/streamEvents.js';
 export * from './public/version.js';

package/dist/index.js

@@ -26,6 +26,7 @@ export * from './public/money.js';
 export * from './public/policy.js';
 export * from './public/providerPreferences.js';
 export * from './public/realtime.js';
+export * from './public/rerank.js';
 export * from './public/responses.js';
 export * from './public/streamEvents.js';
 export * from './public/version.js';

package/dist/internal/catalogue.d.ts

@@ -314,10 +314,10 @@ export declare const routingProfileSchema: z.ZodObject<{
     displayName: z.ZodString;
     description: z.ZodOptional<z.ZodString>;
     optimiseFor: z.ZodEnum<{
+        quality: "quality";
         price: "price";
         latency: "latency";
         throughput: "throughput";
-        quality: "quality";
         balanced: "balanced";
     }>;
     candidates: z.ZodArray<z.ZodObject<{

package/dist/internal/grant.d.ts

@@ -40,6 +40,7 @@ export declare const grantOperationSchema: z.ZodEnum<{
     images_generations: "images_generations";
     audio_transcriptions: "audio_transcriptions";
     audio_speech: "audio_speech";
+    rerank: "rerank";
     realtime_session: "realtime_session";
 }>;
 /**
@@ -101,6 +102,7 @@ export declare const inferenceAuthorizationRequestSchema: z.ZodObject<{
         images_generations: "images_generations";
         audio_transcriptions: "audio_transcriptions";
         audio_speech: "audio_speech";
+        rerank: "rerank";
         realtime_session: "realtime_session";
     }>;
     target: z.ZodOptional<z.ZodDiscriminatedUnion<[z.ZodObject<{
@@ -181,6 +183,7 @@ export declare const authorizationRequestSchema: z.ZodDiscriminatedUnion<[z.ZodO
         images_generations: "images_generations";
         audio_transcriptions: "audio_transcriptions";
         audio_speech: "audio_speech";
+        rerank: "rerank";
         realtime_session: "realtime_session";
     }>;
     target: z.ZodOptional<z.ZodDiscriminatedUnion<[z.ZodObject<{
@@ -383,6 +386,7 @@ export declare const grantSchema: z.ZodObject<{
         images_generations: "images_generations";
         audio_transcriptions: "audio_transcriptions";
         audio_speech: "audio_speech";
+        rerank: "rerank";
         realtime_session: "realtime_session";
     }>;
     issuedAt: z.ZodString;

package/dist/internal/grant.js

@@ -65,6 +65,7 @@ export const grantOperationSchema = z.enum([
     'images_generations',
     'decisions',
     'realtime_session',
+    'rerank',
 ]);
 /**
  * What the caller asked to be served: a model line, optionally pinned to a

package/dist/internal/request.d.ts

@@ -341,6 +341,16 @@ export declare const inferenceRequestSchema: z.ZodObject<{
         }>;
         speed: z.ZodOptional<z.ZodNumber>;
     }, z.core.$strict>>;
+    image: z.ZodOptional<z.ZodObject<{
+        n: z.ZodOptional<z.ZodNumber>;
+        size: z.ZodOptional<z.ZodString>;
+        quality: z.ZodOptional<z.ZodString>;
+        style: z.ZodOptional<z.ZodString>;
+        responseFormat: z.ZodOptional<z.ZodEnum<{
+            url: "url";
+            b64_json: "b64_json";
+        }>>;
+    }, z.core.$strict>>;
     transcription: z.ZodOptional<z.ZodObject<{
         language: z.ZodOptional<z.ZodString>;
         prompt: z.ZodOptional<z.ZodString>;

package/dist/internal/request.js

@@ -8,7 +8,7 @@ import { maxUtf16Length, minUtf16Length } from '../stringBounds.js';
 import { inferenceApiFormatSchema, inferenceModalitySchema } from '../public/catalogue.js';
 import { decisionInputSchema } from '../public/decisions.js';
 import { idempotencyKeySchema, inferenceTimestampSchema } from '../public/identifiers.js';
-import { inferenceAudioOutputParametersSchema, inferenceMessageSchema, inferenceReasoningSchema, inferenceSpeechParametersSchema, inferenceTranscriptionParametersSchema, responseFormatSchema, samplingParametersSchema, toolChoiceSchema, toolDefinitionSchema, } from '../public/messages.js';
+import { inferenceAudioOutputParametersSchema, inferenceMessageSchema, inferenceReasoningSchema, inferenceImageParametersSchema, inferenceSpeechParametersSchema, inferenceTranscriptionParametersSchema, responseFormatSchema, samplingParametersSchema, toolChoiceSchema, toolDefinitionSchema, } from '../public/messages.js';
 import { routingPolicyReferenceSchema } from '../public/policy.js';
 import { inferenceAttributionSchema } from './attribution.js';
 import { authorizedRouteSchema, routingTargetSchema } from './routing.js';
@@ -93,6 +93,8 @@ const inferenceRequestObjectSchema = z.object({
     /** Absent means the route's own default reasoning behaviour. */
     reasoning: inferenceReasoningSchema.optional(),
     speech: inferenceSpeechParametersSchema.optional(),
+    /** Text-to-image controls. Contract set 8.12.0. */
+    image: inferenceImageParametersSchema.optional(),
     /** File transcription parameters. Contract set 4.5.0. */
     transcription: inferenceTranscriptionParametersSchema.optional(),
     /** Spoken output from a conversational model. Contract set 3.2.0. */
@@ -168,6 +170,7 @@ export const inferenceRequestSchema = inferenceRequestObjectSchema.superRefine((
             request.responseFormat !== undefined ||
             request.reasoning !== undefined ||
             request.speech !== undefined ||
+            request.image !== undefined ||
             request.transcription !== undefined ||
             request.audioOutput !== undefined) {
             ctx.addIssue({
@@ -196,6 +199,16 @@ export const inferenceRequestSchema = inferenceRequestObjectSchema.superRefine((
             message: 'speech parameters require the audio_speech API format',
         });
     }
+    if (request.image !== undefined &&
+        (request.client.apiFormat !== 'images_generations' ||
+            request.modality !== 'image' ||
+            request.input.format !== 'text')) {
+        ctx.addIssue({
+            code: 'custom',
+            path: ['image'],
+            message: 'image parameters require the images_generations API format with image modality and text input',
+        });
+    }
     if (request.transcription !== undefined &&
         (request.client.apiFormat !== 'audio_transcriptions' ||
             request.modality !== 'audio' ||

package/dist/public/messages.d.ts

@@ -343,6 +343,24 @@ export declare const inferenceSpeechParametersSchema: z.ZodObject<{
     }>;
     speed: z.ZodOptional<z.ZodNumber>;
 }, z.core.$strict>;
+/**
+ * Parameters for text-to-image (contract set 8.12.0), preserved in the request.
+ * Each is the caller's own control, forwarded as named: an adapter whose
+ * provider takes none of them refuses the request rather than dropping it.
+ * `n` is how many images; `size`, `quality` and `style` are the provider's own
+ * vocabulary (each size/quality class is its own model reference in the
+ * catalogue, so none of them widens a hold).
+ */
+export declare const inferenceImageParametersSchema: z.ZodObject<{
+    n: z.ZodOptional<z.ZodNumber>;
+    size: z.ZodOptional<z.ZodString>;
+    quality: z.ZodOptional<z.ZodString>;
+    style: z.ZodOptional<z.ZodString>;
+    responseFormat: z.ZodOptional<z.ZodEnum<{
+        url: "url";
+        b64_json: "b64_json";
+    }>>;
+}, z.core.$strict>;
 /**
  * Parameters for a file transcription, preserved in the signed request
  * (contract set 4.5.0). `language` is the caller's statement of what is spoken,
@@ -386,6 +404,7 @@ export declare const inferenceAudioOutputParametersSchema: z.ZodObject<{
  * Moved from the Oxy edge (`packages/api/src/schemas/inferenceEdge.schemas.ts`).
  */
 export declare const requestLabelsSchema: z.ZodRecord<z.ZodString, z.ZodString>;
+export type InferenceImageParameters = z.infer<typeof inferenceImageParametersSchema>;
 export type InferenceSpeechParameters = z.infer<typeof inferenceSpeechParametersSchema>;
 export type InferenceTranscriptionParameters = z.infer<typeof inferenceTranscriptionParametersSchema>;
 export type InferenceAudioOutputParameters = z.infer<typeof inferenceAudioOutputParametersSchema>;

package/dist/public/messages.js

@@ -412,6 +412,23 @@ export const inferenceSpeechParametersSchema = z
     speed: z.number().min(0.25).max(4).optional(),
 })
     .strict();
+/**
+ * Parameters for text-to-image (contract set 8.12.0), preserved in the request.
+ * Each is the caller's own control, forwarded as named: an adapter whose
+ * provider takes none of them refuses the request rather than dropping it.
+ * `n` is how many images; `size`, `quality` and `style` are the provider's own
+ * vocabulary (each size/quality class is its own model reference in the
+ * catalogue, so none of them widens a hold).
+ */
+export const inferenceImageParametersSchema = z
+    .object({
+    n: z.number().int().min(1).max(10).optional(),
+    size: z.string().check(minUtf16Length(1)).check(maxUtf16Length(32)).optional(),
+    quality: z.string().check(minUtf16Length(1)).check(maxUtf16Length(32)).optional(),
+    style: z.string().check(minUtf16Length(1)).check(maxUtf16Length(32)).optional(),
+    responseFormat: z.enum(['url', 'b64_json']).optional(),
+})
+    .strict();
 /**
  * Parameters for a file transcription, preserved in the signed request
  * (contract set 4.5.0). `language` is the caller's statement of what is spoken,

package/dist/public/rerank.d.ts

@@ -0,0 +1,38 @@
+/**
+ * `POST /v1/rerank` — rank documents by relevance to a query.
+ *
+ * New in contract set 8.11.0, CONTRACT ONLY: these shapes and the `rerank` grant
+ * operation are declared so Oxy can support the operation before Kaana serves
+ * it. Kaana does not serve the route yet and emits no `rerank` operation.
+ *
+ * Version 1 takes string documents only. The response carries a relevance score
+ * per document and the token usage; it never carries an amount.
+ */
+import { z } from 'zod';
+/** The most documents one rerank request may carry. */
+export declare const MAX_RERANK_DOCUMENTS = 1000;
+export declare const rerankRequestSchema: z.ZodObject<{
+    model: z.ZodString;
+    query: z.ZodString;
+    documents: z.ZodArray<z.ZodString>;
+    top_n: z.ZodOptional<z.ZodNumber>;
+    return_documents: z.ZodOptional<z.ZodBoolean>;
+    user: z.ZodOptional<z.ZodString>;
+}, z.core.$strict>;
+export declare const rerankResponseSchema: z.ZodObject<{
+    id: z.ZodString;
+    model: z.ZodString;
+    results: z.ZodArray<z.ZodObject<{
+        index: z.ZodNumber;
+        relevance_score: z.ZodNumber;
+        document: z.ZodOptional<z.ZodObject<{
+            text: z.ZodString;
+        }, z.core.$strict>>;
+    }, z.core.$strict>>;
+    usage: z.ZodObject<{
+        prompt_tokens: z.ZodNumber;
+        total_tokens: z.ZodNumber;
+    }, z.core.$strict>;
+}, z.core.$strict>;
+export type RerankRequest = z.infer<typeof rerankRequestSchema>;
+export type RerankResponse = z.infer<typeof rerankResponseSchema>;

package/dist/public/rerank.js

@@ -0,0 +1,63 @@
+/**
+ * `POST /v1/rerank` — rank documents by relevance to a query.
+ *
+ * New in contract set 8.11.0, CONTRACT ONLY: these shapes and the `rerank` grant
+ * operation are declared so Oxy can support the operation before Kaana serves
+ * it. Kaana does not serve the route yet and emits no `rerank` operation.
+ *
+ * Version 1 takes string documents only. The response carries a relevance score
+ * per document and the token usage; it never carries an amount.
+ */
+import { z } from 'zod';
+import { maxUtf16Length, minUtf16Length } from '../stringBounds.js';
+import { generationIdSchema, modelReferenceSchema } from './identifiers.js';
+/** The most documents one rerank request may carry. */
+export const MAX_RERANK_DOCUMENTS = 1000;
+export const rerankRequestSchema = z
+    .object({
+    /** `publisher/model` or `publisher/model@revision`. */
+    model: modelReferenceSchema,
+    query: z.string().check(minUtf16Length(1)),
+    /** String documents only. */
+    documents: z
+        .array(z.string().check(minUtf16Length(1)))
+        .min(1)
+        .max(MAX_RERANK_DOCUMENTS),
+    /** How many of the most relevant documents to return; all when absent. */
+    top_n: z.number().int().min(1).optional(),
+    /** Echo each ranked document's text in its result. */
+    return_documents: z.boolean().optional(),
+    /** The delegated end-user id, attribution only (as on `/v1/embeddings`). */
+    user: z.string().check(minUtf16Length(1)).check(maxUtf16Length(64)).optional(),
+})
+    .strict();
+export const rerankResponseSchema = z
+    .object({
+    id: generationIdSchema,
+    /** The revision that produced the scores. */
+    model: modelReferenceSchema,
+    /** Most relevant first. */
+    results: z
+        .array(z
+        .object({
+        /** The position of the document in the request's `documents`. */
+        index: z.number().int().nonnegative(),
+        relevance_score: z.number(),
+        document: z.object({ text: z.string() }).strict().optional(),
+    })
+        .strict())
+        .max(MAX_RERANK_DOCUMENTS),
+    usage: z
+        .object({
+        prompt_tokens: z.number().int().nonnegative(),
+        total_tokens: z.number().int().nonnegative(),
+    })
+        .strict(),
+})
+    .strict()
+    .superRefine((response, ctx) => {
+    const indexes = response.results.map((result) => result.index);
+    if (new Set(indexes).size !== indexes.length) {
+        ctx.addIssue({ code: 'custom', path: ['results'], message: 'result indexes must be unique' });
+    }
+});

package/dist/public/version.d.ts

@@ -100,7 +100,7 @@
  * change to, say, the catalogue reject every in-flight inference request; the
  * per-shape `schemaVersion` is what a message is validated against.
  */
-export declare const INFERENCE_CONTRACT_VERSION = "8.10.0";
+export declare const INFERENCE_CONTRACT_VERSION = "8.12.0";
 /**
  * Contract set 4.0.0 (Kaana ADR 0002) is the first published as
  * `@kaana.ai/contracts`. MAJOR because the set changed hands and split in two:
@@ -236,6 +236,18 @@ export declare const INFERENCE_CONTRACT_VERSION = "8.10.0";
  * Both read the declared list price of a model line, in USD per million tokens.
  * Public request members only: the signed grant and envelope are unchanged and
  * no response carries an amount.
+ *
+ * Contract set 8.11.0 is additive and CONTRACT ONLY: the `rerank` grant
+ * operation (`grantOperationSchema`, last member) and the public `POST /v1/rerank`
+ * shapes (`rerankRequestSchema`, `rerankResponseSchema`). Kaana emits no `rerank`
+ * operation and serves no route yet; Oxy supports the operation first.
+ *
+ * Contract set 8.12.0 is additive: the signed envelope's optional `image`
+ * member (`inferenceImageParametersSchema`: `n`, `size`, `quality`, `style`,
+ * `responseFormat`), which carries a text-to-image request's controls to the
+ * adapter instead of dropping them. It is valid only with the
+ * `images_generations` API format, image modality and text input. Nothing in
+ * 8.11.0 changes and the public request and response shapes are unchanged.
  */
 /** The public API's one origin and version. */
 export declare const KAANA_API_BASE_URL = "https://kaana.ai/v1";

package/dist/public/version.js

@@ -100,7 +100,7 @@
  * change to, say, the catalogue reject every in-flight inference request; the
  * per-shape `schemaVersion` is what a message is validated against.
  */
-export const INFERENCE_CONTRACT_VERSION = '8.10.0';
+export const INFERENCE_CONTRACT_VERSION = '8.12.0';
 /**
  * Contract set 4.0.0 (Kaana ADR 0002) is the first published as
  * `@kaana.ai/contracts`. MAJOR because the set changed hands and split in two:
@@ -236,6 +236,18 @@ export const INFERENCE_CONTRACT_VERSION = '8.10.0';
  * Both read the declared list price of a model line, in USD per million tokens.
  * Public request members only: the signed grant and envelope are unchanged and
  * no response carries an amount.
+ *
+ * Contract set 8.11.0 is additive and CONTRACT ONLY: the `rerank` grant
+ * operation (`grantOperationSchema`, last member) and the public `POST /v1/rerank`
+ * shapes (`rerankRequestSchema`, `rerankResponseSchema`). Kaana emits no `rerank`
+ * operation and serves no route yet; Oxy supports the operation first.
+ *
+ * Contract set 8.12.0 is additive: the signed envelope's optional `image`
+ * member (`inferenceImageParametersSchema`: `n`, `size`, `quality`, `style`,
+ * `responseFormat`), which carries a text-to-image request's controls to the
+ * adapter instead of dropping them. It is valid only with the
+ * `images_generations` API format, image modality and text input. Nothing in
+ * 8.11.0 changes and the public request and response shapes are unchanged.
  */
 /** The public API's one origin and version. */
 export const KAANA_API_BASE_URL = 'https://kaana.ai/v1';

package/openapi.json

@@ -2,7 +2,7 @@
   "openapi": "3.1.0",
   "info": {
     "title": "Kaana API",
-    "version": "8.10.0",
+    "version": "8.12.0",
     "description": "Kaana inference API. Generated from @kaana.ai/contracts; the Zod source is the authority. Accounts, applications, oxy_sk_ keys, limits and billing are managed in Oxy Console: https://console.oxy.so"
   },
   "servers": [
@@ -4311,6 +4311,39 @@
         "maxLength": 2048,
         "pattern": "^https:\\/\\/[^\\s]+$"
       },
+      "InferenceImageParameters": {
+        "type": "object",
+        "properties": {
+          "n": {
+            "type": "integer",
+            "minimum": 1,
+            "maximum": 10
+          },
+          "size": {
+            "type": "string",
+            "minLength": 1,
+            "maxLength": 32
+          },
+          "quality": {
+            "type": "string",
+            "minLength": 1,
+            "maxLength": 32
+          },
+          "style": {
+            "type": "string",
+            "minLength": 1,
+            "maxLength": 32
+          },
+          "responseFormat": {
+            "type": "string",
+            "enum": [
+              "url",
+              "b64_json"
+            ]
+          }
+        },
+        "additionalProperties": false
+      },
       "InferenceMessageRole": {
         "type": "string",
         "enum": [
@@ -7718,6 +7751,118 @@
           "maxLength": 256
         }
       },
+      "RerankRequest": {
+        "type": "object",
+        "properties": {
+          "model": {
+            "$ref": "#/components/schemas/ModelReference"
+          },
+          "query": {
+            "type": "string",
+            "minLength": 1
+          },
+          "documents": {
+            "minItems": 1,
+            "maxItems": 1000,
+            "type": "array",
+            "items": {
+              "type": "string",
+              "minLength": 1
+            }
+          },
+          "top_n": {
+            "type": "integer",
+            "minimum": 1,
+            "maximum": 9007199254740991
+          },
+          "return_documents": {
+            "type": "boolean"
+          },
+          "user": {
+            "type": "string",
+            "minLength": 1,
+            "maxLength": 64
+          }
+        },
+        "required": [
+          "model",
+          "query",
+          "documents"
+        ],
+        "additionalProperties": false
+      },
+      "RerankResponse": {
+        "type": "object",
+        "properties": {
+          "id": {
+            "$ref": "#/components/schemas/GenerationId"
+          },
+          "model": {
+            "$ref": "#/components/schemas/ModelReference"
+          },
+          "results": {
+            "maxItems": 1000,
+            "type": "array",
+            "items": {
+              "type": "object",
+              "properties": {
+                "index": {
+                  "type": "integer",
+                  "minimum": 0,
+                  "maximum": 9007199254740991
+                },
+                "relevance_score": {
+                  "type": "number"
+                },
+                "document": {
+                  "type": "object",
+                  "properties": {
+                    "text": {
+                      "type": "string"
+                    }
+                  },
+                  "required": [
+                    "text"
+                  ],
+                  "additionalProperties": false
+                }
+              },
+              "required": [
+                "index",
+                "relevance_score"
+              ],
+              "additionalProperties": false
+            }
+          },
+          "usage": {
+            "type": "object",
+            "properties": {
+              "prompt_tokens": {
+                "type": "integer",
+                "minimum": 0,
+                "maximum": 9007199254740991
+              },
+              "total_tokens": {
+                "type": "integer",
+                "minimum": 0,
+                "maximum": 9007199254740991
+              }
+            },
+            "required": [
+              "prompt_tokens",
+              "total_tokens"
+            ],
+            "additionalProperties": false
+          }
+        },
+        "required": [
+          "id",
+          "model",
+          "results",
+          "usage"
+        ],
+        "additionalProperties": false
+      },
       "ResponseFormat": {
         "oneOf": [
           {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@kaana.ai/contracts",
-  "version": "8.10.0",
+  "version": "8.12.0",
   "description": "Kaana inference contract: the Zod schemas of the public API at https://kaana.ai/v1 and, under /internal, of the Oxy-Kaana grant, settlement and signed lane.",
   "type": "module",
   "sideEffects": false,