@kaana.ai/contracts 5.0.0 → 6.0.0

package/dist/public/catalogue.d.ts

@@ -116,120 +116,6 @@ export declare const realtimeSessionKindSchema: z.ZodEnum<{
 export declare const realtimeSessionTransportSchema: z.ZodEnum<{
     websocket: "websocket";
 }>;
-/** A closed interval. */
-export declare const numberRangeSchema: z.ZodObject<{
-    min: z.ZodNumber;
-    max: z.ZodNumber;
-}, z.core.$strict>;
-/** One kind of tool choice: the modes, and `function` for one that names a tool. */
-export declare const realtimeToolChoiceKindSchema: z.ZodEnum<{
-    function: "function";
-    none: "none";
-    auto: "auto";
-    required: "required";
-}>;
-/**
- * One turn detection a realtime model serves (contract set 5.0.0).
- *
- * - `fixedResponses` — the detection always answers the turn it detected and
- *   lets the caller's speech interrupt the answer: it has neither switch, so a
- *   session must state `createResponse: true` and `interruptResponse: true`.
- * - `threshold` — server_vad's `threshold` range; absent, the contract's 0–1.
- * - `meteredUnits` — the units a session under this detection is billed in:
- *   what a caller multiplies the model's prices by to estimate a call. A
- *   provider that bills by the clock under one detection and by the audio
- *   under another states each.
- */
-export declare const realtimeTurnDetectionParametersSchema: z.ZodObject<{
-    type: z.ZodEnum<{
-        none: "none";
-        server_vad: "server_vad";
-        semantic_vad: "semantic_vad";
-    }>;
-    fixedResponses: z.ZodBoolean;
-    threshold: z.ZodOptional<z.ZodObject<{
-        min: z.ZodNumber;
-        max: z.ZodNumber;
-    }, z.core.$strict>>;
-    meteredUnits: z.ZodArray<z.ZodEnum<{
-        input_tokens: "input_tokens";
-        cached_input_tokens: "cached_input_tokens";
-        output_tokens: "output_tokens";
-        reasoning_tokens: "reasoning_tokens";
-        requests: "requests";
-        images: "images";
-        audio_input_milliseconds: "audio_input_milliseconds";
-        audio_output_milliseconds: "audio_output_milliseconds";
-        video_milliseconds: "video_milliseconds";
-        characters: "characters";
-        embeddings: "embeddings";
-        audio_input_tokens: "audio_input_tokens";
-        cached_audio_input_tokens: "cached_audio_input_tokens";
-        audio_output_tokens: "audio_output_tokens";
-        session_milliseconds: "session_milliseconds";
-        input_transcription_milliseconds: "input_transcription_milliseconds";
-    }>>;
-}, z.core.$strict>;
-/**
- * What a realtime model's session configuration may set, and within which
- * bounds (contract set 5.0.0): the provider's reviewed statement, which Kaana's
- * adapter enforces and the catalogue publishes, so a session built from it is
- * one the model serves. `checkRealtimeSessionOpen` applies it.
- *
- * - `turnDetection` — every detection served.
- * - `outputModalities` — every set of modalities one response may produce
- *   (each sorted): a model that answers aloud whatever it is asked does not
- *   serve `["text"]`.
- * - `toolChoice` — every kind of tool choice served; always holds `auto`.
- * - `maxOutputTokens` — the largest per-response cap; absent, the model has
- *   none and `maxOutputTokens` is refused.
- * - `inputTranscriptionPrompt` — `config.inputAudioTranscription.prompt` is
- *   taken.
- */
-export declare const realtimeParametersSchema: z.ZodObject<{
-    turnDetection: z.ZodArray<z.ZodObject<{
-        type: z.ZodEnum<{
-            none: "none";
-            server_vad: "server_vad";
-            semantic_vad: "semantic_vad";
-        }>;
-        fixedResponses: z.ZodBoolean;
-        threshold: z.ZodOptional<z.ZodObject<{
-            min: z.ZodNumber;
-            max: z.ZodNumber;
-        }, z.core.$strict>>;
-        meteredUnits: z.ZodArray<z.ZodEnum<{
-            input_tokens: "input_tokens";
-            cached_input_tokens: "cached_input_tokens";
-            output_tokens: "output_tokens";
-            reasoning_tokens: "reasoning_tokens";
-            requests: "requests";
-            images: "images";
-            audio_input_milliseconds: "audio_input_milliseconds";
-            audio_output_milliseconds: "audio_output_milliseconds";
-            video_milliseconds: "video_milliseconds";
-            characters: "characters";
-            embeddings: "embeddings";
-            audio_input_tokens: "audio_input_tokens";
-            cached_audio_input_tokens: "cached_audio_input_tokens";
-            audio_output_tokens: "audio_output_tokens";
-            session_milliseconds: "session_milliseconds";
-            input_transcription_milliseconds: "input_transcription_milliseconds";
-        }>>;
-    }, z.core.$strict>>;
-    outputModalities: z.ZodArray<z.ZodArray<z.ZodEnum<{
-        text: "text";
-        audio: "audio";
-    }>>>;
-    toolChoice: z.ZodArray<z.ZodEnum<{
-        function: "function";
-        none: "none";
-        auto: "auto";
-        required: "required";
-    }>>;
-    maxOutputTokens: z.ZodOptional<z.ZodNumber>;
-    inputTranscriptionPrompt: z.ZodBoolean;
-}, z.core.$strict>;
 /**
  * What a deployment can execute as a realtime session. Present only on a model
  * that holds sessions; absent means it holds none, which is the state of every
@@ -247,8 +133,6 @@ export declare const realtimeParametersSchema: z.ZodObject<{
  * - `audioFormats` — the realtime encodings a session may name as
  *   `config.inputAudioFormat` and `config.outputAudioFormat`. These are not the
  *   media types of `audio.audioFormats`, which describe files.
- * - `parameters` (contract set 5.0.0) — what a session configuration may set
- *   ({@link realtimeParametersSchema}). Absent: undeclared, not checked.
  */
 export declare const modelRealtimeCapabilitiesSchema: z.ZodObject<{
     transports: z.ZodArray<z.ZodEnum<{
@@ -272,50 +156,6 @@ export declare const modelRealtimeCapabilitiesSchema: z.ZodObject<{
             g711_alaw: "g711_alaw";
         }>>;
     }, z.core.$strict>>;
-    parameters: z.ZodOptional<z.ZodObject<{
-        turnDetection: z.ZodArray<z.ZodObject<{
-            type: z.ZodEnum<{
-                none: "none";
-                server_vad: "server_vad";
-                semantic_vad: "semantic_vad";
-            }>;
-            fixedResponses: z.ZodBoolean;
-            threshold: z.ZodOptional<z.ZodObject<{
-                min: z.ZodNumber;
-                max: z.ZodNumber;
-            }, z.core.$strict>>;
-            meteredUnits: z.ZodArray<z.ZodEnum<{
-                input_tokens: "input_tokens";
-                cached_input_tokens: "cached_input_tokens";
-                output_tokens: "output_tokens";
-                reasoning_tokens: "reasoning_tokens";
-                requests: "requests";
-                images: "images";
-                audio_input_milliseconds: "audio_input_milliseconds";
-                audio_output_milliseconds: "audio_output_milliseconds";
-                video_milliseconds: "video_milliseconds";
-                characters: "characters";
-                embeddings: "embeddings";
-                audio_input_tokens: "audio_input_tokens";
-                cached_audio_input_tokens: "cached_audio_input_tokens";
-                audio_output_tokens: "audio_output_tokens";
-                session_milliseconds: "session_milliseconds";
-                input_transcription_milliseconds: "input_transcription_milliseconds";
-            }>>;
-        }, z.core.$strict>>;
-        outputModalities: z.ZodArray<z.ZodArray<z.ZodEnum<{
-            text: "text";
-            audio: "audio";
-        }>>>;
-        toolChoice: z.ZodArray<z.ZodEnum<{
-            function: "function";
-            none: "none";
-            auto: "auto";
-            required: "required";
-        }>>;
-        maxOutputTokens: z.ZodOptional<z.ZodNumber>;
-        inputTranscriptionPrompt: z.ZodBoolean;
-    }, z.core.$strict>>;
 }, z.core.$strict>;
 /**
  * What a model can do, in the terms a caller has to decide against before
@@ -385,50 +225,6 @@ export declare const modelCapabilitiesSchema: z.ZodObject<{
                 g711_alaw: "g711_alaw";
             }>>;
         }, z.core.$strict>>;
-        parameters: z.ZodOptional<z.ZodObject<{
-            turnDetection: z.ZodArray<z.ZodObject<{
-                type: z.ZodEnum<{
-                    none: "none";
-                    server_vad: "server_vad";
-                    semantic_vad: "semantic_vad";
-                }>;
-                fixedResponses: z.ZodBoolean;
-                threshold: z.ZodOptional<z.ZodObject<{
-                    min: z.ZodNumber;
-                    max: z.ZodNumber;
-                }, z.core.$strict>>;
-                meteredUnits: z.ZodArray<z.ZodEnum<{
-                    input_tokens: "input_tokens";
-                    cached_input_tokens: "cached_input_tokens";
-                    output_tokens: "output_tokens";
-                    reasoning_tokens: "reasoning_tokens";
-                    requests: "requests";
-                    images: "images";
-                    audio_input_milliseconds: "audio_input_milliseconds";
-                    audio_output_milliseconds: "audio_output_milliseconds";
-                    video_milliseconds: "video_milliseconds";
-                    characters: "characters";
-                    embeddings: "embeddings";
-                    audio_input_tokens: "audio_input_tokens";
-                    cached_audio_input_tokens: "cached_audio_input_tokens";
-                    audio_output_tokens: "audio_output_tokens";
-                    session_milliseconds: "session_milliseconds";
-                    input_transcription_milliseconds: "input_transcription_milliseconds";
-                }>>;
-            }, z.core.$strict>>;
-            outputModalities: z.ZodArray<z.ZodArray<z.ZodEnum<{
-                text: "text";
-                audio: "audio";
-            }>>>;
-            toolChoice: z.ZodArray<z.ZodEnum<{
-                function: "function";
-                none: "none";
-                auto: "auto";
-                required: "required";
-            }>>;
-            maxOutputTokens: z.ZodOptional<z.ZodNumber>;
-            inputTranscriptionPrompt: z.ZodBoolean;
-        }, z.core.$strict>>;
     }, z.core.$strict>>;
 }, z.core.$strict>;
 /**
@@ -557,15 +353,11 @@ export declare const voiceSchema: z.ZodObject<{
  * `gender` is published only where it has been reviewed against the
  * provider's own catalogue; absent is unknown.
  *
- * Contract set 5.0.0:
- *
- * - `detectsLanguage` — naming no language lets the model detect it. A caller
- *   whose language is not in `languages` omits it rather than losing the
- *   model; absent is undeclared.
- * - `transcriptionPrompt` — a transcription request may carry a `prompt`.
- *   Absent or `false`: one that does is refused before anything is held.
- * - A TRANSCRIPTION model's `languages` are exactly what a request may name in
- *   `language`, and its `audioFormats.input` exactly what it uploads.
+ * A model's `languages` describe it; they are not a list a request must pick
+ * from. A request names the client's own BCP 47 tag, and Kaana sends the
+ * provider its spelling of that language, or none for the model to detect
+ * (Kaana ADR 0003). A transcription model's `audioFormats.input` are the
+ * uploads it reads.
  */
 export declare const modelAudioCapabilitiesSchema: z.ZodObject<{
     voices: z.ZodArray<z.ZodObject<{
@@ -579,8 +371,6 @@ export declare const modelAudioCapabilitiesSchema: z.ZodObject<{
         }>>;
     }, z.core.$strict>>;
     languages: z.ZodArray<z.ZodString>;
-    detectsLanguage: z.ZodOptional<z.ZodBoolean>;
-    transcriptionPrompt: z.ZodOptional<z.ZodBoolean>;
     audioFormats: z.ZodObject<{
         input: z.ZodArray<z.ZodEnum<{
             "audio/mpeg": "audio/mpeg";
@@ -692,50 +482,6 @@ export declare const modelSchema: z.ZodObject<{
                     g711_alaw: "g711_alaw";
                 }>>;
             }, z.core.$strict>>;
-            parameters: z.ZodOptional<z.ZodObject<{
-                turnDetection: z.ZodArray<z.ZodObject<{
-                    type: z.ZodEnum<{
-                        none: "none";
-                        server_vad: "server_vad";
-                        semantic_vad: "semantic_vad";
-                    }>;
-                    fixedResponses: z.ZodBoolean;
-                    threshold: z.ZodOptional<z.ZodObject<{
-                        min: z.ZodNumber;
-                        max: z.ZodNumber;
-                    }, z.core.$strict>>;
-                    meteredUnits: z.ZodArray<z.ZodEnum<{
-                        input_tokens: "input_tokens";
-                        cached_input_tokens: "cached_input_tokens";
-                        output_tokens: "output_tokens";
-                        reasoning_tokens: "reasoning_tokens";
-                        requests: "requests";
-                        images: "images";
-                        audio_input_milliseconds: "audio_input_milliseconds";
-                        audio_output_milliseconds: "audio_output_milliseconds";
-                        video_milliseconds: "video_milliseconds";
-                        characters: "characters";
-                        embeddings: "embeddings";
-                        audio_input_tokens: "audio_input_tokens";
-                        cached_audio_input_tokens: "cached_audio_input_tokens";
-                        audio_output_tokens: "audio_output_tokens";
-                        session_milliseconds: "session_milliseconds";
-                        input_transcription_milliseconds: "input_transcription_milliseconds";
-                    }>>;
-                }, z.core.$strict>>;
-                outputModalities: z.ZodArray<z.ZodArray<z.ZodEnum<{
-                    text: "text";
-                    audio: "audio";
-                }>>>;
-                toolChoice: z.ZodArray<z.ZodEnum<{
-                    function: "function";
-                    none: "none";
-                    auto: "auto";
-                    required: "required";
-                }>>;
-                maxOutputTokens: z.ZodOptional<z.ZodNumber>;
-                inputTranscriptionPrompt: z.ZodBoolean;
-            }, z.core.$strict>>;
         }, z.core.$strict>>;
     }, z.core.$strict>;
     audio: z.ZodOptional<z.ZodObject<{
@@ -750,8 +496,6 @@ export declare const modelSchema: z.ZodObject<{
             }>>;
         }, z.core.$strict>>;
         languages: z.ZodArray<z.ZodString>;
-        detectsLanguage: z.ZodOptional<z.ZodBoolean>;
-        transcriptionPrompt: z.ZodOptional<z.ZodBoolean>;
         audioFormats: z.ZodObject<{
             input: z.ZodArray<z.ZodEnum<{
                 "audio/mpeg": "audio/mpeg";
@@ -835,6 +579,10 @@ export declare const modelSchema: z.ZodObject<{
             currency: z.ZodString;
         }, z.core.$strict>>;
     }, z.core.$strict>>;
+    pricePerMinute: z.ZodOptional<z.ZodObject<{
+        amount: z.core.$ZodBranded<z.ZodString, "ExactDecimal", "out">;
+        currency: z.ZodString;
+    }, z.core.$strict>>;
     deprecation: z.ZodObject<{
         status: z.ZodEnum<{
             active: "active";
@@ -953,50 +701,6 @@ export declare const modelListSchema: z.ZodObject<{
                         g711_alaw: "g711_alaw";
                     }>>;
                 }, z.core.$strict>>;
-                parameters: z.ZodOptional<z.ZodObject<{
-                    turnDetection: z.ZodArray<z.ZodObject<{
-                        type: z.ZodEnum<{
-                            none: "none";
-                            server_vad: "server_vad";
-                            semantic_vad: "semantic_vad";
-                        }>;
-                        fixedResponses: z.ZodBoolean;
-                        threshold: z.ZodOptional<z.ZodObject<{
-                            min: z.ZodNumber;
-                            max: z.ZodNumber;
-                        }, z.core.$strict>>;
-                        meteredUnits: z.ZodArray<z.ZodEnum<{
-                            input_tokens: "input_tokens";
-                            cached_input_tokens: "cached_input_tokens";
-                            output_tokens: "output_tokens";
-                            reasoning_tokens: "reasoning_tokens";
-                            requests: "requests";
-                            images: "images";
-                            audio_input_milliseconds: "audio_input_milliseconds";
-                            audio_output_milliseconds: "audio_output_milliseconds";
-                            video_milliseconds: "video_milliseconds";
-                            characters: "characters";
-                            embeddings: "embeddings";
-                            audio_input_tokens: "audio_input_tokens";
-                            cached_audio_input_tokens: "cached_audio_input_tokens";
-                            audio_output_tokens: "audio_output_tokens";
-                            session_milliseconds: "session_milliseconds";
-                            input_transcription_milliseconds: "input_transcription_milliseconds";
-                        }>>;
-                    }, z.core.$strict>>;
-                    outputModalities: z.ZodArray<z.ZodArray<z.ZodEnum<{
-                        text: "text";
-                        audio: "audio";
-                    }>>>;
-                    toolChoice: z.ZodArray<z.ZodEnum<{
-                        function: "function";
-                        none: "none";
-                        auto: "auto";
-                        required: "required";
-                    }>>;
-                    maxOutputTokens: z.ZodOptional<z.ZodNumber>;
-                    inputTranscriptionPrompt: z.ZodBoolean;
-                }, z.core.$strict>>;
             }, z.core.$strict>>;
         }, z.core.$strict>;
         audio: z.ZodOptional<z.ZodObject<{
@@ -1011,8 +715,6 @@ export declare const modelListSchema: z.ZodObject<{
                 }>>;
             }, z.core.$strict>>;
             languages: z.ZodArray<z.ZodString>;
-            detectsLanguage: z.ZodOptional<z.ZodBoolean>;
-            transcriptionPrompt: z.ZodOptional<z.ZodBoolean>;
             audioFormats: z.ZodObject<{
                 input: z.ZodArray<z.ZodEnum<{
                     "audio/mpeg": "audio/mpeg";
@@ -1096,6 +798,10 @@ export declare const modelListSchema: z.ZodObject<{
                 currency: z.ZodString;
             }, z.core.$strict>>;
         }, z.core.$strict>>;
+        pricePerMinute: z.ZodOptional<z.ZodObject<{
+            amount: z.core.$ZodBranded<z.ZodString, "ExactDecimal", "out">;
+            currency: z.ZodString;
+        }, z.core.$strict>>;
         deprecation: z.ZodObject<{
             status: z.ZodEnum<{
                 active: "active";
@@ -1206,9 +912,6 @@ export type InferenceApiFormat = z.infer<typeof inferenceApiFormatSchema>;
 export type RealtimeSessionKind = z.infer<typeof realtimeSessionKindSchema>;
 export type RealtimeSessionTransport = z.infer<typeof realtimeSessionTransportSchema>;
 export type ModelRealtimeCapabilities = z.infer<typeof modelRealtimeCapabilitiesSchema>;
-export type RealtimeParameters = z.infer<typeof realtimeParametersSchema>;
-export type RealtimeTurnDetectionParameters = z.infer<typeof realtimeTurnDetectionParametersSchema>;
-export type RealtimeToolChoiceKind = z.infer<typeof realtimeToolChoiceKindSchema>;
 export type ModelCapabilities = z.infer<typeof modelCapabilitiesSchema>;
 export type ModelLicense = z.infer<typeof modelLicenseSchema>;
 export type ModelProvenance = z.infer<typeof modelProvenanceSchema>;

package/dist/public/catalogue.js

@@ -5,8 +5,8 @@
  */
 import { z } from 'zod';
 import { inferenceDateSchema, inferenceHttpsUrlSchema, inferenceRegionSchema, inferenceTimestampSchema, modelIdSchema, modelReferenceSchema, modelRevisionLabelSchema, publisherSlugSchema, } from './identifiers.js';
-import { audioInputMediaTypeSchema, languageTagSchema, realtimeAudioFormatSchema, realtimeOutputModalitySchema, realtimeTurnDetectionTypeSchema, } from './audioFormats.js';
-import { priceSnapshotSchema, usageUnitSchema } from './money.js';
+import { audioInputMediaTypeSchema, languageTagSchema, realtimeAudioFormatSchema, } from './audioFormats.js';
+import { moneySchema, priceSnapshotSchema } from './money.js';
 import { inferenceAudioMediaTypeSchema } from './streamEvents.js';
 /** Input/request modalities; structured decisions are an output, never an input. */
 export const inferenceModalitySchema = z.enum(['text', 'image', 'audio', 'video', 'embedding']);
@@ -94,68 +94,6 @@ export const realtimeSessionKindSchema = z.enum(['conversation', 'transcription'
  * deployment can actually execute them.
  */
 export const realtimeSessionTransportSchema = z.enum(['websocket']);
-/** A closed interval. */
-export const numberRangeSchema = z.object({ min: z.number(), max: z.number() }).strict();
-/** One kind of tool choice: the modes, and `function` for one that names a tool. */
-export const realtimeToolChoiceKindSchema = z.enum(['auto', 'none', 'required', 'function']);
-/**
- * One turn detection a realtime model serves (contract set 5.0.0).
- *
- * - `fixedResponses` — the detection always answers the turn it detected and
- *   lets the caller's speech interrupt the answer: it has neither switch, so a
- *   session must state `createResponse: true` and `interruptResponse: true`.
- * - `threshold` — server_vad's `threshold` range; absent, the contract's 0–1.
- * - `meteredUnits` — the units a session under this detection is billed in:
- *   what a caller multiplies the model's prices by to estimate a call. A
- *   provider that bills by the clock under one detection and by the audio
- *   under another states each.
- */
-export const realtimeTurnDetectionParametersSchema = z
-    .object({
-    type: realtimeTurnDetectionTypeSchema,
-    fixedResponses: z.boolean(),
-    threshold: numberRangeSchema.optional(),
-    meteredUnits: z.array(usageUnitSchema).min(1),
-})
-    .strict();
-/**
- * What a realtime model's session configuration may set, and within which
- * bounds (contract set 5.0.0): the provider's reviewed statement, which Kaana's
- * adapter enforces and the catalogue publishes, so a session built from it is
- * one the model serves. `checkRealtimeSessionOpen` applies it.
- *
- * - `turnDetection` — every detection served.
- * - `outputModalities` — every set of modalities one response may produce
- *   (each sorted): a model that answers aloud whatever it is asked does not
- *   serve `["text"]`.
- * - `toolChoice` — every kind of tool choice served; always holds `auto`.
- * - `maxOutputTokens` — the largest per-response cap; absent, the model has
- *   none and `maxOutputTokens` is refused.
- * - `inputTranscriptionPrompt` — `config.inputAudioTranscription.prompt` is
- *   taken.
- */
-export const realtimeParametersSchema = z
-    .object({
-    turnDetection: z.array(realtimeTurnDetectionParametersSchema).min(1),
-    outputModalities: z.array(z.array(realtimeOutputModalitySchema).min(1).max(2)).min(1),
-    toolChoice: z.array(realtimeToolChoiceKindSchema).min(1),
-    maxOutputTokens: z.number().int().positive().safe().optional(),
-    inputTranscriptionPrompt: z.boolean(),
-})
-    .strict()
-    .superRefine((parameters, ctx) => {
-    if (!parameters.toolChoice.includes('auto')) {
-        ctx.addIssue({ code: 'custom', path: ['toolChoice'], message: 'auto is always served' });
-    }
-    const kinds = parameters.turnDetection.map((detection) => detection.type);
-    if (new Set(kinds).size !== kinds.length) {
-        ctx.addIssue({
-            code: 'custom',
-            path: ['turnDetection'],
-            message: 'each detection is declared once',
-        });
-    }
-});
 /**
  * What a deployment can execute as a realtime session. Present only on a model
  * that holds sessions; absent means it holds none, which is the state of every
@@ -173,8 +111,6 @@ export const realtimeParametersSchema = z
  * - `audioFormats` — the realtime encodings a session may name as
  *   `config.inputAudioFormat` and `config.outputAudioFormat`. These are not the
  *   media types of `audio.audioFormats`, which describe files.
- * - `parameters` (contract set 5.0.0) — what a session configuration may set
- *   ({@link realtimeParametersSchema}). Absent: undeclared, not checked.
  */
 export const modelRealtimeCapabilitiesSchema = z
     .object({
@@ -188,7 +124,6 @@ export const modelRealtimeCapabilitiesSchema = z
     })
         .strict()
         .optional(),
-    parameters: realtimeParametersSchema.optional(),
 })
     .strict();
 /**
@@ -444,22 +379,16 @@ export const voiceSchema = z
  * `gender` is published only where it has been reviewed against the
  * provider's own catalogue; absent is unknown.
  *
- * Contract set 5.0.0:
- *
- * - `detectsLanguage` — naming no language lets the model detect it. A caller
- *   whose language is not in `languages` omits it rather than losing the
- *   model; absent is undeclared.
- * - `transcriptionPrompt` — a transcription request may carry a `prompt`.
- *   Absent or `false`: one that does is refused before anything is held.
- * - A TRANSCRIPTION model's `languages` are exactly what a request may name in
- *   `language`, and its `audioFormats.input` exactly what it uploads.
+ * A model's `languages` describe it; they are not a list a request must pick
+ * from. A request names the client's own BCP 47 tag, and Kaana sends the
+ * provider its spelling of that language, or none for the model to detect
+ * (Kaana ADR 0003). A transcription model's `audioFormats.input` are the
+ * uploads it reads.
  */
 export const modelAudioCapabilitiesSchema = z
     .object({
     voices: z.array(voiceSchema),
     languages: z.array(languageTagSchema),
-    detectsLanguage: z.boolean().optional(),
-    transcriptionPrompt: z.boolean().optional(),
     audioFormats: z
         .object({
         input: z.array(audioInputMediaTypeSchema),
@@ -546,6 +475,14 @@ export const modelSchema = z.object({
     dataPolicy: inferenceDataPolicySchema,
     /** Oxy's list price. Absent for a model a customer cannot buy per unit (BYOK-only). */
     pricing: priceSnapshotSchema.optional(),
+    /**
+     * Kaana's price of one minute of this model's work, whatever units its
+     * provider bills (contract set 6.0.0): for a realtime model, a minute of a
+     * conversation under `server_vad`, Kaana's default turn detection; for a
+     * transcription model, a minute of audio. An estimate to reserve or show
+     * with, rounded up; what a request costs is what it measured, in `pricing`.
+     */
+    pricePerMinute: moneySchema.optional(),
     deprecation: modelDeprecationSchema,
     evaluations: z.array(modelEvaluationResultSchema).default([]),
     safety: modelSafetyMetadataSchema.optional(),

package/dist/public/chatCompletions.d.ts

@@ -172,8 +172,8 @@ export declare const chatCompletionsRequestSchema: z.ZodObject<{
         }, z.core.$strict>;
     }, z.core.$strict>>>;
     tool_choice: z.ZodOptional<z.ZodUnion<readonly [z.ZodEnum<{
-        none: "none";
         auto: "auto";
+        none: "none";
         required: "required";
     }>, z.ZodObject<{
         type: z.ZodLiteral<"function">;

package/dist/public/messages.d.ts

@@ -205,8 +205,8 @@ export declare const toolDefinitionSchema: z.ZodObject<{
 }, z.core.$strict>;
 /** Whether, and which, tool the model must call. */
 export declare const toolChoiceSchema: z.ZodUnion<readonly [z.ZodEnum<{
-    none: "none";
     auto: "auto";
+    none: "none";
     required: "required";
 }>, z.ZodObject<{
     type: z.ZodLiteral<"function">;