@juspay/neurolink 11.27.0 → 11.29.0

package/dist/cli/factories/commandFactory.d.ts

@@ -300,6 +300,7 @@ export declare class CLICommandFactory {
         };
         videoProvider: {
             type: "string";
+            choices: string[];
             description: string;
         };
         videoOutput: {
@@ -328,6 +329,7 @@ export declare class CLICommandFactory {
         };
         avatarProvider: {
             type: "string";
+            choices: string[];
             description: string;
         };
         avatarImage: {
@@ -362,6 +364,7 @@ export declare class CLICommandFactory {
         };
         musicProvider: {
             type: "string";
+            choices: string[];
             description: string;
         };
         musicDuration: {
@@ -1190,6 +1193,7 @@ export declare const commonOptions: {
     };
     videoProvider: {
         type: "string";
+        choices: string[];
         description: string;
     };
     videoOutput: {
@@ -1218,6 +1222,7 @@ export declare const commonOptions: {
     };
     avatarProvider: {
         type: "string";
+        choices: string[];
         description: string;
     };
     avatarImage: {
@@ -1252,6 +1257,7 @@ export declare const commonOptions: {
     };
     musicProvider: {
         type: "string";
+        choices: string[];
         description: string;
     };
     musicDuration: {

package/dist/cli/factories/commandFactory.js

@@ -3,6 +3,7 @@ import path from "node:path";
 import chalk from "chalk";
 import ora from "ora";
 import { ModelResolver } from "../../models/modelResolver.js";
+import { providerChoicesFor } from "../../factories/mediaHandlerCatalog.js";
 import { globalSession } from "../../session/globalSessionState.js";
 // Use TokenUsage from standard types - no local interface needed
 import { ContextFactory, } from "../../types/index.js";
@@ -333,7 +334,7 @@ export class CLICommandFactory {
         },
         ttsProvider: {
             type: "string",
-            choices: ["google-ai", "vertex", "openai-tts", "elevenlabs", "azure-tts"],
+            choices: providerChoicesFor("tts"),
             description: "TTS provider (overrides --provider for speech synthesis)",
         },
         ttsFormat: {
@@ -381,7 +382,7 @@ export class CLICommandFactory {
         },
         sttProvider: {
             type: "string",
-            choices: ["whisper", "deepgram", "google-stt", "azure-stt"],
+            choices: providerChoicesFor("stt"),
             description: "STT provider to use",
         },
         sttLanguage: {
@@ -401,6 +402,7 @@ export class CLICommandFactory {
         },
         videoProvider: {
             type: "string",
+            choices: providerChoicesFor("video"),
             description: "Video provider override (e.g., 'vertex' (default), 'kling', 'runway', 'replicate')",
         },
         videoOutput: {
@@ -430,6 +432,7 @@ export class CLICommandFactory {
         // Avatar Generation options (D-ID, HeyGen, MuseTalk via Replicate)
         avatarProvider: {
             type: "string",
+            choices: providerChoicesFor("avatar"),
             description: "Avatar provider (e.g., 'd-id' (default), 'heygen', 'replicate', 'musetalk')",
         },
         avatarImage: {
@@ -465,6 +468,7 @@ export class CLICommandFactory {
         // Music Generation options (Beatoven, ElevenLabs, Lyria, MusicGen via Replicate)
         musicProvider: {
             type: "string",
+            choices: providerChoicesFor("music"),
             description: "Music provider (e.g., 'beatoven' (default), 'elevenlabs-music', 'lyria', 'replicate', 'musicgen')",
         },
         musicDuration: {

package/dist/core/baseProvider.js

@@ -1,8 +1,9 @@
 import { context, SpanKind, SpanStatusCode, trace } from "@opentelemetry/api";
 import { directAgentTools } from "../agent/directTools.js";
-import { isImageGenerationModel } from "./constants.js";
+import { defaultProviderFor } from "../factories/mediaHandlerCatalog.js";
 import { MiddlewareFactory } from "../middleware/factory.js";
 import { modelSupports } from "../models/modelRegistry.js";
+import { resolveRequestKind } from "./resolveRequestKind.js";
 import { ATTR, tracers } from "../telemetry/index.js";
 import { ERROR_CODES, isAbortError, NeuroLinkError, } from "../utils/errorHandling.js";
 import { ProviderError } from "../types/index.js";
@@ -238,14 +239,12 @@ export class BaseProvider {
         }
         // CRITICAL: Image generation models don't support real streaming
         // Force fake streaming for image models to ensure image output is yielded.
-        // Skip this path when the caller explicitly requests non-image output (e.g.
-        // JSON analysis) so dual-mode models like gemini-3.1-flash-image-preview
-        // can still perform text/structured generation.
-        const isImageModel = isImageGenerationModel(this.modelName);
-        const requestsNonImageOutput = options.output?.format === "json" ||
-            options.output?.format === "structured" ||
-            options.output?.format === "text";
-        if (isImageModel && !requestsNonImageOutput) {
+        // resolveRequestKind() skips this path when the caller explicitly requests
+        // non-image output (e.g. JSON analysis) so dual-mode models like
+        // gemini-3.1-flash-image-preview can still perform text/structured
+        // generation — see its doc comment for the full precedence table.
+        const requestKind = resolveRequestKind(options, this.modelName);
+        if (requestKind === "image") {
             logger.info(`Image model detected, forcing fake streaming`, {
                 provider: this.providerName,
                 model: this.modelName,
@@ -1066,14 +1065,13 @@ export class BaseProvider {
     }
     async runGenerateInActiveContext(options, startTime, otelSpan, otelSpanState) {
         try {
-            if (options.output?.mode === "video") {
+            // Single source of truth for "what kind of request is this" — see
+            // resolveRequestKind's doc comment for the full precedence table.
+            const requestKind = resolveRequestKind(options, this.modelName);
+            if (requestKind === "video") {
                 return await this.handleVideoGeneration(options, startTime);
             }
-            const isImageModel = isImageGenerationModel(this.modelName);
-            const requestsNonImageOutput = options.output?.format === "json" ||
-                options.output?.format === "structured" ||
-                options.output?.format === "text";
-            if (isImageModel && !requestsNonImageOutput) {
+            if (requestKind === "image") {
                 logger.info(`Image generation model detected, routing to executeImageGeneration`, {
                     provider: this.providerName,
                     model: this.modelName,
@@ -1081,7 +1079,7 @@ export class BaseProvider {
                 const imageResult = await this.executeImageGeneration(options);
                 return await this.enhanceResult(imageResult, options, startTime);
             }
-            if (options.tts?.enabled && !options.tts?.useAiResponse) {
+            if (requestKind === "tts-direct") {
                 return this.handleDirectTTSSynthesis(options, startTime);
             }
             const { tools, model } = await this.prepareGenerationContext(options);
@@ -2139,9 +2137,10 @@ export class BaseProvider {
         }
         // Get prompt text
         const prompt = options.prompt || options.input?.text || "";
-        // Honor output.video.provider — when omitted, fall back to "vertex"
-        // for backward compatibility with the original implementation.
-        const requestedProvider = options.output?.video?.provider ?? "vertex";
+        // Honor output.video.provider — when omitted, fall back to the
+        // catalog-derived default (currently "vertex") for backward
+        // compatibility with the original implementation.
+        const requestedProvider = options.output?.video?.provider ?? defaultProviderFor("video");
         if (!VideoProcessor.supports(requestedProvider)) {
             throw new VideoError({
                 code: VIDEO_ERROR_CODES.PROVIDER_NOT_SUPPORTED,
@@ -2177,7 +2176,12 @@ export class BaseProvider {
         // shared timeout helper so standard video gen honors the caller's
         // timeout the same way director mode does (see above ~Line 2062).
         const videoTimeout = options.timeout ?? 600_000; // 10 min default
-        const videoResult = await this.executeWithTimeout(() => VideoProcessor.generate(requestedProvider, imageBuffer, prompt, options.output?.video ?? {}, options.region), { timeout: videoTimeout, operationType: "generate" });
+        const videoResult = await this.executeWithTimeout(() => VideoProcessor.generate(requestedProvider, {
+            ...(options.output?.video ?? {}),
+            image: imageBuffer,
+            prompt,
+            region: options.region,
+        }), { timeout: videoTimeout, operationType: "generate" });
         // Prefer the handler's own model id (more accurate — it knows the exact
         // checkpoint that ran). Fall back to the request-time value, and finally
         // to the Vertex default only when we're on the Vertex route.

package/dist/core/resolveRequestKind.d.ts

@@ -0,0 +1,27 @@
+import type { RequestKind, RequestKindInput } from "../types/index.js";
+/**
+ * The dispatch decision for "what kind of request is this" — text, image,
+ * video, music, avatar, direct TTS synthesis, or PPT generation — at the
+ * CORE call sites: neurolink.ts's maybeHandleEarlyGenerateResult
+ * (music/avatar/ppt/workflow routing) and baseProvider.ts's
+ * stream()/runGenerateInActiveContext (image/video/tts-direct routing) call
+ * this instead of independently re-deriving the decision.
+ *
+ * NOT yet the only copy: replicate.ts's generate() override and
+ * googleVertex/client.ts's native dispatch (~6672-6718) still carry their
+ * own provider-internal versions — the Vertex one with a cruder
+ * startsWith() image match. Migrating those two is queued follow-up work;
+ * until it lands, an edit to this precedence table does not reach them.
+ *
+ * Precedence, checked in order:
+ *   1. output.mode (music/avatar/video/ppt) — an explicit mode always wins.
+ *   2. an image-generation model, unless the caller explicitly asked for a
+ *      non-image output.format (json/structured/text) — this lets dual-mode
+ *      models like gemini-3.1-flash-image-preview still perform text or
+ *      structured generation when requested.
+ *   3. tts.enabled without tts.useAiResponse — direct synthesis, bypassing
+ *      the LLM turn entirely (useAiResponse means the LLM's own text
+ *      response gets synthesized afterward, which is NOT this branch).
+ *   4. otherwise, "text".
+ */
+export declare function resolveRequestKind(options: RequestKindInput, modelName?: string): RequestKind;

package/dist/core/resolveRequestKind.js

@@ -0,0 +1,50 @@
+import { isImageGenerationModel } from "./constants.js";
+/**
+ * The dispatch decision for "what kind of request is this" — text, image,
+ * video, music, avatar, direct TTS synthesis, or PPT generation — at the
+ * CORE call sites: neurolink.ts's maybeHandleEarlyGenerateResult
+ * (music/avatar/ppt/workflow routing) and baseProvider.ts's
+ * stream()/runGenerateInActiveContext (image/video/tts-direct routing) call
+ * this instead of independently re-deriving the decision.
+ *
+ * NOT yet the only copy: replicate.ts's generate() override and
+ * googleVertex/client.ts's native dispatch (~6672-6718) still carry their
+ * own provider-internal versions — the Vertex one with a cruder
+ * startsWith() image match. Migrating those two is queued follow-up work;
+ * until it lands, an edit to this precedence table does not reach them.
+ *
+ * Precedence, checked in order:
+ *   1. output.mode (music/avatar/video/ppt) — an explicit mode always wins.
+ *   2. an image-generation model, unless the caller explicitly asked for a
+ *      non-image output.format (json/structured/text) — this lets dual-mode
+ *      models like gemini-3.1-flash-image-preview still perform text or
+ *      structured generation when requested.
+ *   3. tts.enabled without tts.useAiResponse — direct synthesis, bypassing
+ *      the LLM turn entirely (useAiResponse means the LLM's own text
+ *      response gets synthesized afterward, which is NOT this branch).
+ *   4. otherwise, "text".
+ */
+export function resolveRequestKind(options, modelName) {
+    if (options.output?.mode === "music") {
+        return "music";
+    }
+    if (options.output?.mode === "avatar") {
+        return "avatar";
+    }
+    if (options.output?.mode === "video") {
+        return "video";
+    }
+    if (options.output?.mode === "ppt") {
+        return "ppt";
+    }
+    const requestsNonImageOutput = options.output?.format === "json" ||
+        options.output?.format === "structured" ||
+        options.output?.format === "text";
+    if (isImageGenerationModel(modelName) && !requestsNonImageOutput) {
+        return "image";
+    }
+    if (options.tts?.enabled && !options.tts?.useAiResponse) {
+        return "tts-direct";
+    }
+    return "text";
+}

package/dist/factories/mediaHandlerCatalog.d.ts

@@ -0,0 +1,18 @@
+import type { MediaHandlerDescriptor, MediaHandlerKind } from "../types/index.js";
+/**
+ * Static catalog of every shipped media-handler provider, across all six
+ * ecosystems (TTS, STT, Realtime, Video, Avatar, Music). Pure data — no
+ * factory functions, no class imports — mirroring the
+ * src/lib/factories/providerDescriptors.ts pattern for text/image
+ * providers.
+ *
+ * Entries mirror providerRegistry.ts's hand-constructed TTS/STT/Realtime/
+ * Video/Avatar/Music registration blocks (registerAllProviders()) exactly —
+ * this file states today's truth, not an aspirational shape. Keep it in
+ * sync if those blocks change.
+ */
+export declare const MEDIA_HANDLER_CATALOG: readonly MediaHandlerDescriptor[];
+/** Every selectable provider name for `kind`, primaries and aliases both. */
+export declare function providerChoicesFor(kind: MediaHandlerKind): string[];
+/** The first-listed primary provider name for `kind` — used as a fallback default. */
+export declare function defaultProviderFor(kind: MediaHandlerKind): string;

package/dist/factories/mediaHandlerCatalog.js

@@ -0,0 +1,65 @@
+/**
+ * Static catalog of every shipped media-handler provider, across all six
+ * ecosystems (TTS, STT, Realtime, Video, Avatar, Music). Pure data — no
+ * factory functions, no class imports — mirroring the
+ * src/lib/factories/providerDescriptors.ts pattern for text/image
+ * providers.
+ *
+ * Entries mirror providerRegistry.ts's hand-constructed TTS/STT/Realtime/
+ * Video/Avatar/Music registration blocks (registerAllProviders()) exactly —
+ * this file states today's truth, not an aspirational shape. Keep it in
+ * sync if those blocks change.
+ */
+export const MEDIA_HANDLER_CATALOG = [
+    // --- TTS ---
+    { kind: "tts", name: "google-ai", aliases: ["vertex"] },
+    { kind: "tts", name: "openai-tts" },
+    { kind: "tts", name: "elevenlabs", aliases: ["elevenlabs-tts"] },
+    { kind: "tts", name: "azure-tts" },
+    { kind: "tts", name: "fish-audio" },
+    { kind: "tts", name: "cartesia" },
+    // --- STT ---
+    { kind: "stt", name: "whisper", aliases: ["openai-stt"] },
+    { kind: "stt", name: "deepgram" },
+    { kind: "stt", name: "google-stt" },
+    { kind: "stt", name: "azure-stt" },
+    // --- Realtime ---
+    { kind: "realtime", name: "openai-realtime" },
+    { kind: "realtime", name: "gemini-live" },
+    // --- Video ---
+    { kind: "video", name: "vertex" },
+    { kind: "video", name: "kling" },
+    { kind: "video", name: "runway" },
+    { kind: "video", name: "replicate" },
+    // --- Avatar ---
+    { kind: "avatar", name: "d-id" },
+    { kind: "avatar", name: "replicate", aliases: ["musetalk"] },
+    { kind: "avatar", name: "heygen" },
+    // --- Music ---
+    { kind: "music", name: "beatoven" },
+    { kind: "music", name: "replicate", aliases: ["musicgen"] },
+    { kind: "music", name: "elevenlabs-music", aliases: ["elevenlabs-sound"] },
+    { kind: "music", name: "lyria" },
+];
+/** Every selectable provider name for `kind`, primaries and aliases both. */
+export function providerChoicesFor(kind) {
+    const choices = [];
+    for (const entry of MEDIA_HANDLER_CATALOG) {
+        if (entry.kind !== kind) {
+            continue;
+        }
+        choices.push(entry.name);
+        if (entry.aliases) {
+            choices.push(...entry.aliases);
+        }
+    }
+    return choices;
+}
+/** The first-listed primary provider name for `kind` — used as a fallback default. */
+export function defaultProviderFor(kind) {
+    const first = MEDIA_HANDLER_CATALOG.find((entry) => entry.kind === kind);
+    if (!first) {
+        throw new Error(`No media handler catalog entries registered for kind "${kind}"`);
+    }
+    return first.name;
+}