@juspay/neurolink 13.1.1 → 13.2.0

package/dist/types/providerCatalog.generated.d.ts

@@ -1,2 +1,2 @@
-export type CatalogProviderName = "a2agent" | "abliteration" | "above-dev" | "ai21" | "aiand" | "aihubmix" | "aionlabs" | "ambient" | "api-route" | "arcee" | "atlas-cloud" | "avian-io" | "baidu-qianfan" | "baseten" | "bee-heossi" | "byteplus-modelark" | "bytez" | "cerebras" | "charm-hyper" | "chutes" | "cloudflare" | "crusoe" | "dashscope" | "deepinfra" | "deepseek" | "empiriolabs" | "featherless-ai" | "fireworks" | "friendli" | "gmicloud" | "gradientai" | "groq" | "hetzner-inference" | "huggingface" | "inception-labs" | "inceptron" | "inco" | "inference-net" | "io-intelligence" | "koscompute" | "laya" | "lemonfox-ai" | "lilac" | "llmtech" | "longcat" | "mancer" | "meta-model-api" | "minimax" | "mistral" | "moark" | "modelscope" | "moonshot-ai" | "morph" | "naver-clova-studio" | "nebius" | "neuralwatt" | "novita" | "openai-decisions" | "ovhcloud" | "parasail" | "pareto-inference" | "perplexity" | "poolside" | "prime-intellect" | "reka" | "runinfra" | "sakana-ai" | "sambanova" | "sarvam" | "scaleway" | "siliconflow" | "stackit" | "stepfun" | "subconscious" | "subq" | "synthetic" | "telnyx" | "thinking-machines" | "tinfoil" | "together-ai" | "umans-ai" | "upstage" | "venice-ai" | "vispark" | "vultr-inference" | "wafer" | "wandb-inference" | "xai" | "xiaomi-mimo" | "xor" | "z-ai";
-export type CatalogCredentialKey = "a2agent" | "abliteration" | "aboveDev" | "ai21" | "aiand" | "aihubmix" | "aionlabs" | "ambient" | "apiRoute" | "arcee" | "atlasCloud" | "avianIo" | "baiduQianfan" | "baseten" | "beeHeossi" | "byteplusModelark" | "bytez" | "cerebras" | "charmHyper" | "chutes" | "cloudflare" | "crusoe" | "dashscope" | "deepinfra" | "deepseek" | "empiriolabs" | "featherlessAi" | "fireworks" | "friendli" | "gmicloud" | "gradientai" | "groq" | "hetznerInference" | "huggingFace" | "inceptionLabs" | "inceptron" | "inco" | "inferenceNet" | "ioIntelligence" | "koscompute" | "laya" | "lemonfoxAi" | "lilac" | "llmtech" | "longcat" | "mancer" | "metaModelApi" | "minimax" | "mistral" | "moark" | "modelscope" | "moonshotAi" | "morph" | "naverClovaStudio" | "nebius" | "neuralwatt" | "novita" | "openaiDecisions" | "ovhcloud" | "parasail" | "paretoInference" | "perplexity" | "poolside" | "primeIntellect" | "reka" | "runinfra" | "sakanaAi" | "sambanova" | "sarvam" | "scaleway" | "siliconflow" | "stackit" | "stepfun" | "subconscious" | "subq" | "synthetic" | "telnyx" | "thinkingMachines" | "tinfoil" | "together" | "umansAi" | "upstage" | "veniceAi" | "vispark" | "vultrInference" | "wafer" | "wandbInference" | "xai" | "xiaomiMimo" | "xor" | "zAi";
+export type CatalogProviderName = "a2agent" | "abliteration" | "above-dev" | "ai21" | "aiand" | "aihubmix" | "aionlabs" | "ambient" | "api-route" | "arcee" | "atlas-cloud" | "avian-io" | "baidu-qianfan" | "baseten" | "bee-heossi" | "byteplus-modelark" | "bytez" | "cerebras" | "charm-hyper" | "chutes" | "cloudflare" | "crusoe" | "dashscope" | "deepinfra" | "deepseek" | "empiriolabs" | "featherless-ai" | "fireworks" | "friendli" | "gmicloud" | "gradientai" | "groq" | "hetzner-inference" | "huggingface" | "inception-labs" | "inceptron" | "inco" | "inference-net" | "io-intelligence" | "koscompute" | "laya" | "lemonfox-ai" | "lilac" | "llmtech" | "longcat" | "mancer" | "meta-model-api" | "minimax" | "mistral" | "moark" | "modelscope" | "moonshot-ai" | "morph" | "naver-clova-studio" | "nebius" | "neuralwatt" | "novita" | "openai-decisions" | "ovhcloud" | "parasail" | "pareto-inference" | "perplexity" | "poolside" | "prime-intellect" | "reka" | "runinfra" | "sakana-ai" | "sambanova" | "sarvam" | "scaledown" | "scaleway" | "siliconflow" | "stackit" | "stepfun" | "subconscious" | "subq" | "synthetic" | "telnyx" | "thinking-machines" | "tinfoil" | "together-ai" | "umans-ai" | "upstage" | "venice-ai" | "vispark" | "vultr-inference" | "wafer" | "wandb-inference" | "xai" | "xiaomi-mimo" | "xor" | "z-ai";
+export type CatalogCredentialKey = "a2agent" | "abliteration" | "aboveDev" | "ai21" | "aiand" | "aihubmix" | "aionlabs" | "ambient" | "apiRoute" | "arcee" | "atlasCloud" | "avianIo" | "baiduQianfan" | "baseten" | "beeHeossi" | "byteplusModelark" | "bytez" | "cerebras" | "charmHyper" | "chutes" | "cloudflare" | "crusoe" | "dashscope" | "deepinfra" | "deepseek" | "empiriolabs" | "featherlessAi" | "fireworks" | "friendli" | "gmicloud" | "gradientai" | "groq" | "hetznerInference" | "huggingFace" | "inceptionLabs" | "inceptron" | "inco" | "inferenceNet" | "ioIntelligence" | "koscompute" | "laya" | "lemonfoxAi" | "lilac" | "llmtech" | "longcat" | "mancer" | "metaModelApi" | "minimax" | "mistral" | "moark" | "modelscope" | "moonshotAi" | "morph" | "naverClovaStudio" | "nebius" | "neuralwatt" | "novita" | "openaiDecisions" | "ovhcloud" | "parasail" | "paretoInference" | "perplexity" | "poolside" | "primeIntellect" | "reka" | "runinfra" | "sakanaAi" | "sambanova" | "sarvam" | "scaledown" | "scaleway" | "siliconflow" | "stackit" | "stepfun" | "subconscious" | "subq" | "synthetic" | "telnyx" | "thinkingMachines" | "tinfoil" | "together" | "umansAi" | "upstage" | "veniceAi" | "vispark" | "vultrInference" | "wafer" | "wandbInference" | "xai" | "xiaomiMimo" | "xor" | "zAi";

package/dist/types/providers.d.ts

@@ -469,6 +469,10 @@ export type NeurolinkCredentials = {
         apiKey?: string;
         baseURL?: string;
     };
+    scaledown?: {
+        apiKey?: string;
+        baseURL?: string;
+    };
     scaleway?: {
         apiKey?: string;
         baseURL?: string;

package/dist/utils/pricing.js

@@ -812,6 +812,18 @@ const PRICING = {
         // name, so a `"gpt-6-luna"` key would price that chat model at this rate.
         _default: { input: 0.1 / 1_000_000, output: 0 },
     },
+    scaledown: {
+        // $0.05 per million input tokens, output free: `output: 0` is a price, not
+        // a token count, so do not "correct" it from `usage.output_tokens`. A list
+        // price from ScaleDown's launch post and from `usage.cost` in a response,
+        // both as reported to this project on 2026-10-15; this project has not
+        // priced a call itself. Per ScaleDown's docs each choice option is a
+        // separate internal call, so input tokens grow with options times state
+        // length (not measured here). Only `_default`: LiteLLM, OpenRouter and
+        // openai-compatible search every table by model name, so a `"classify"` key
+        // would price an unrelated model at this rate.
+        _default: { input: 0.05 / 1_000_000, output: 0 },
+    },
     "cloudflare-clef": {
         // Workers AI bills Clef per million INPUT tokens ($0.24 for clef, $0.09 for
         // clef-flash); no output price is listed. `output: 0` is a price, not a token

package/dist/utils/tokenLimits.js

@@ -2,7 +2,7 @@
  * Provider-specific token limit utilities
  * Provides safe maxTokens values based on provider and model capabilities
  */
-import { getRuntimeOutputCeiling } from "../constants/contextWindows.js";
+import { getCatalogOutputCeiling, getRuntimeOutputCeiling, } from "../constants/contextWindows.js";
 import { bedrockModelIdForLookup, stripBedrockGeoPrefix, } from "../constants/bedrockIdentifiers.js";
 import { PROVIDER_MAX_TOKENS } from "../core/constants.js";
 import { logger } from "./logger.js";
@@ -68,6 +68,25 @@ export function getSafeMaxTokens(provider, model, requestedMaxTokens) {
     // Get provider-specific limits
     const providerLimits = PROVIDER_MAX_TOKENS[canonicalProvider];
     if (!providerLimits) {
+        // A JSON-catalog provider has no PROVIDER_MAX_TOKENS row, but its catalog
+        // entry can declare the model's own output ceiling. Only that per-model
+        // value counts: the provider-wide default is a 4096 placeholder for many
+        // providers. An over-large request is clamped. Unlike the runtime ceiling
+        // above, this ceiling is not injected as a default: an unset request stays
+        // unset, because a ceiling-sized default would shrink usable input on
+        // total-context backends.
+        const catalogCeiling = getCatalogOutputCeiling(provider, model);
+        if (catalogCeiling !== undefined) {
+            if (requestedMaxTokens === undefined || requestedMaxTokens === null) {
+                return undefined;
+            }
+            if (requestedMaxTokens > catalogCeiling) {
+                logger.warn(`Requested maxTokens ${requestedMaxTokens} exceeds the catalog output ceiling ` +
+                    `of ${catalogCeiling} for ${provider}/${model}. Using ${catalogCeiling} instead.`);
+                return catalogCeiling;
+            }
+            return requestedMaxTokens;
+        }
         logger.warn(`Unknown provider ${provider}, no token limits enforced`);
         // Explicit undefined/null check so a caller-supplied 0 is preserved
         // (truthy guard would silently drop 0 here as it does in the restricted branch).