@dexto/core 1.4.0 → 1.5.0

package/dist/llm/registry.cjs

@@ -36,8 +36,11 @@ __export(registry_exports, {
   getSupportedProviders: () => getSupportedProviders,
   isValidProviderModel: () => isValidProviderModel,
   modelSupportsFileType: () => modelSupportsFileType,
+  requiresApiKey: () => requiresApiKey,
   requiresBaseURL: () => requiresBaseURL,
+  stripBedrockRegionPrefix: () => stripBedrockRegionPrefix,
   supportsBaseURL: () => supportsBaseURL,
+  supportsCustomModels: () => supportsCustomModels,
   validateModelFileSupport: () => validateModelFileSupport
 });
 module.exports = __toCommonJS(registry_exports);
@@ -45,6 +48,7 @@ var import_errors = require("./errors.js");
 var import_error_codes = require("./error-codes.js");
 var import_DextoRuntimeError = require("../errors/DextoRuntimeError.js");
 var import_types = require("./types.js");
+var import_openrouter_model_registry = require("./providers/openrouter-model-registry.js");
 const MIME_TYPE_TO_FILE_TYPE = {
   "application/pdf": "pdf",
   "audio/mp3": "audio",
@@ -274,6 +278,7 @@ const LLM_REGISTRY = {
         pricing: {
           inputPerM: 2.5,
           outputPerM: 10,
+          cacheReadPerM: 1.25,
           currency: "USD",
           unit: "per_million_tokens"
         }
@@ -339,8 +344,9 @@ const LLM_REGISTRY = {
     models: [],
     // Empty - accepts any model name for custom endpoints
     baseURLSupport: "required",
-    supportedFileTypes: ["pdf", "image", "audio"]
+    supportedFileTypes: ["pdf", "image", "audio"],
     // Allow all types for custom endpoints - user assumes responsibility for model capabilities
+    supportsCustomModels: true
   },
   anthropic: {
     models: [
@@ -478,6 +484,20 @@ const LLM_REGISTRY = {
   },
   google: {
     models: [
+      {
+        name: "gemini-3-flash-preview",
+        displayName: "Gemini 3 Flash Preview",
+        maxInputTokens: 1048576,
+        default: true,
+        supportedFileTypes: ["pdf", "image", "audio"],
+        pricing: {
+          inputPerM: 0.5,
+          outputPerM: 3,
+          cacheReadPerM: 0.05,
+          currency: "USD",
+          unit: "per_million_tokens"
+        }
+      },
       {
         name: "gemini-3-pro-preview",
         displayName: "Gemini 3 Pro Preview",
@@ -486,6 +506,7 @@ const LLM_REGISTRY = {
         pricing: {
           inputPerM: 2,
           outputPerM: 12,
+          cacheReadPerM: 0.2,
           currency: "USD",
           unit: "per_million_tokens"
         }
@@ -498,6 +519,7 @@ const LLM_REGISTRY = {
         pricing: {
           inputPerM: 2,
           outputPerM: 120,
+          cacheReadPerM: 0.2,
           currency: "USD",
           unit: "per_million_tokens"
         }
@@ -506,7 +528,6 @@ const LLM_REGISTRY = {
         name: "gemini-2.5-pro",
         displayName: "Gemini 2.5 Pro",
         maxInputTokens: 1048576,
-        default: true,
         supportedFileTypes: ["pdf", "image", "audio"],
         pricing: {
           inputPerM: 1.25,
@@ -524,6 +545,7 @@ const LLM_REGISTRY = {
         pricing: {
           inputPerM: 0.3,
           outputPerM: 2.5,
+          cacheReadPerM: 0.03,
           currency: "USD",
           unit: "per_million_tokens"
         }
@@ -563,6 +585,7 @@ const LLM_REGISTRY = {
         pricing: {
           inputPerM: 0.075,
           outputPerM: 0.3,
+          cacheReadPerM: 0.01875,
           currency: "USD",
           unit: "per_million_tokens"
         }
@@ -619,6 +642,7 @@ const LLM_REGISTRY = {
         pricing: {
           inputPerM: 1,
           outputPerM: 3,
+          cacheReadPerM: 0.5,
           currency: "USD",
           unit: "per_million_tokens"
         }
@@ -807,8 +831,462 @@ const LLM_REGISTRY = {
     baseURLSupport: "none",
     supportedFileTypes: []
     // Cohere currently doesn't support file uploads
+  },
+  // https://openrouter.ai/docs
+  // OpenRouter is a unified API gateway providing access to 100+ models from various providers.
+  // Model validation is handled dynamically via openrouter-model-registry.ts
+  openrouter: {
+    models: [],
+    // Empty - accepts any model name (validated against OpenRouter's catalog)
+    baseURLSupport: "none",
+    // Fixed endpoint - baseURL auto-injected in resolver, no user override allowed
+    supportedFileTypes: ["pdf", "image", "audio"],
+    // Allow all types - user assumes responsibility for model capabilities
+    supportsCustomModels: true
+  },
+  // https://docs.litellm.ai/
+  // LiteLLM is an OpenAI-compatible proxy that unifies 100+ LLM providers.
+  // User must host their own LiteLLM proxy and provide the baseURL.
+  litellm: {
+    models: [],
+    // Empty - accepts any model name (user's proxy determines available models)
+    baseURLSupport: "required",
+    // User must provide their LiteLLM proxy URL
+    supportedFileTypes: ["pdf", "image", "audio"],
+    // Allow all types - user assumes responsibility for model capabilities
+    supportsCustomModels: true
+  },
+  // https://glama.ai/
+  // Glama is an OpenAI-compatible gateway providing unified access to multiple LLM providers.
+  // Fixed endpoint: https://glama.ai/api/gateway/openai/v1
+  glama: {
+    models: [],
+    // Empty - accepts any model name (format: provider/model e.g., openai/gpt-4o)
+    baseURLSupport: "none",
+    // Fixed endpoint - baseURL auto-injected
+    supportedFileTypes: ["pdf", "image", "audio"],
+    // Allow all types - user assumes responsibility for model capabilities
+    supportsCustomModels: true
+  },
+  // https://cloud.google.com/vertex-ai
+  // Google Vertex AI - GCP-hosted gateway for Gemini and Claude models
+  // Supports both Google's Gemini models and Anthropic's Claude via partnership
+  //
+  // Setup instructions:
+  // 1. Create a Google Cloud account and project
+  // 2. Enable the Vertex AI API: gcloud services enable aiplatform.googleapis.com
+  // 3. Enable desired Claude models (requires Anthropic Model Garden)
+  // 4. Install Google Cloud CLI: https://cloud.google.com/sdk/docs/install
+  // 5. Configure ADC: gcloud auth application-default login
+  // 6. Set env vars: GOOGLE_VERTEX_PROJECT (required), GOOGLE_VERTEX_LOCATION (optional)
+  //
+  // TODO: Add dynamic model fetching via publishers.models.list API
+  // - Requires: projectId, region, ADC auth
+  // - Endpoints: GET projects/{project}/locations/{location}/publishers/{google,anthropic}/models
+  // - Note: API doesn't return aliases (e.g., gemini-2.0-flash), only versioned IDs
+  // - Docs: https://cloud.google.com/vertex-ai/docs/reference/rest/v1/projects.locations.models/list
+  // - Models: https://cloud.google.com/vertex-ai/generative-ai/docs/models
+  vertex: {
+    models: [
+      // Gemini 3 models on Vertex AI (Preview)
+      {
+        name: "gemini-3-flash-preview",
+        displayName: "Gemini 3 Flash (Vertex)",
+        maxInputTokens: 1048576,
+        default: true,
+        supportedFileTypes: ["pdf", "image", "audio"],
+        pricing: {
+          inputPerM: 0.5,
+          outputPerM: 3,
+          cacheReadPerM: 0.05,
+          currency: "USD",
+          unit: "per_million_tokens"
+        }
+      },
+      {
+        name: "gemini-3-pro-preview",
+        displayName: "Gemini 3 Pro (Vertex)",
+        maxInputTokens: 1048576,
+        supportedFileTypes: ["pdf", "image", "audio"],
+        pricing: {
+          inputPerM: 2,
+          outputPerM: 12,
+          cacheReadPerM: 0.2,
+          currency: "USD",
+          unit: "per_million_tokens"
+        }
+      },
+      // Gemini 2.x models on Vertex AI
+      {
+        name: "gemini-2.5-pro",
+        displayName: "Gemini 2.5 Pro (Vertex)",
+        maxInputTokens: 1048576,
+        supportedFileTypes: ["pdf", "image", "audio"],
+        pricing: {
+          inputPerM: 1.25,
+          outputPerM: 10,
+          cacheReadPerM: 0.31,
+          currency: "USD",
+          unit: "per_million_tokens"
+        }
+      },
+      {
+        name: "gemini-2.5-flash",
+        displayName: "Gemini 2.5 Flash (Vertex)",
+        maxInputTokens: 1048576,
+        supportedFileTypes: ["pdf", "image", "audio"],
+        pricing: {
+          inputPerM: 0.15,
+          outputPerM: 0.6,
+          cacheReadPerM: 0.0375,
+          currency: "USD",
+          unit: "per_million_tokens"
+        }
+      },
+      {
+        name: "gemini-2.0-flash",
+        displayName: "Gemini 2.0 Flash (Vertex)",
+        maxInputTokens: 1048576,
+        supportedFileTypes: ["pdf", "image", "audio"],
+        pricing: {
+          inputPerM: 0.1,
+          outputPerM: 0.4,
+          cacheReadPerM: 0.025,
+          currency: "USD",
+          unit: "per_million_tokens"
+        }
+      },
+      // Claude 4.5 models on Vertex AI (via Anthropic partnership)
+      // Note: Claude model IDs use @ suffix format on Vertex
+      {
+        name: "claude-opus-4-5@20251101",
+        displayName: "Claude 4.5 Opus (Vertex)",
+        maxInputTokens: 2e5,
+        supportedFileTypes: ["pdf", "image"],
+        pricing: {
+          inputPerM: 5,
+          outputPerM: 25,
+          cacheWritePerM: 6.25,
+          cacheReadPerM: 0.5,
+          currency: "USD",
+          unit: "per_million_tokens"
+        }
+      },
+      {
+        name: "claude-sonnet-4-5@20250929",
+        displayName: "Claude 4.5 Sonnet (Vertex)",
+        maxInputTokens: 2e5,
+        supportedFileTypes: ["pdf", "image"],
+        pricing: {
+          inputPerM: 3,
+          outputPerM: 15,
+          cacheWritePerM: 3.75,
+          cacheReadPerM: 0.3,
+          currency: "USD",
+          unit: "per_million_tokens"
+        }
+      },
+      {
+        name: "claude-haiku-4-5@20251001",
+        displayName: "Claude 4.5 Haiku (Vertex)",
+        maxInputTokens: 2e5,
+        supportedFileTypes: ["pdf", "image"],
+        pricing: {
+          inputPerM: 1,
+          outputPerM: 5,
+          cacheWritePerM: 1.25,
+          cacheReadPerM: 0.1,
+          currency: "USD",
+          unit: "per_million_tokens"
+        }
+      },
+      // Claude 4.1 and 4.0 models on Vertex AI
+      {
+        name: "claude-opus-4-1@20250805",
+        displayName: "Claude 4.1 Opus (Vertex)",
+        maxInputTokens: 2e5,
+        supportedFileTypes: ["pdf", "image"],
+        pricing: {
+          inputPerM: 15,
+          outputPerM: 75,
+          cacheWritePerM: 18.75,
+          cacheReadPerM: 1.5,
+          currency: "USD",
+          unit: "per_million_tokens"
+        }
+      },
+      {
+        name: "claude-opus-4@20250514",
+        displayName: "Claude 4 Opus (Vertex)",
+        maxInputTokens: 2e5,
+        supportedFileTypes: ["pdf", "image"],
+        pricing: {
+          inputPerM: 15,
+          outputPerM: 75,
+          cacheWritePerM: 18.75,
+          cacheReadPerM: 1.5,
+          currency: "USD",
+          unit: "per_million_tokens"
+        }
+      },
+      {
+        name: "claude-sonnet-4@20250514",
+        displayName: "Claude 4 Sonnet (Vertex)",
+        maxInputTokens: 2e5,
+        supportedFileTypes: ["pdf", "image"],
+        pricing: {
+          inputPerM: 3,
+          outputPerM: 15,
+          cacheWritePerM: 3.75,
+          cacheReadPerM: 0.3,
+          currency: "USD",
+          unit: "per_million_tokens"
+        }
+      },
+      // Claude 3.x models on Vertex AI
+      {
+        name: "claude-3-7-sonnet@20250219",
+        displayName: "Claude 3.7 Sonnet (Vertex)",
+        maxInputTokens: 2e5,
+        supportedFileTypes: ["pdf", "image"],
+        pricing: {
+          inputPerM: 3,
+          outputPerM: 15,
+          cacheWritePerM: 3.75,
+          cacheReadPerM: 0.3,
+          currency: "USD",
+          unit: "per_million_tokens"
+        }
+      },
+      {
+        name: "claude-3-5-sonnet-v2@20241022",
+        displayName: "Claude 3.5 Sonnet v2 (Vertex)",
+        maxInputTokens: 2e5,
+        supportedFileTypes: ["pdf", "image"],
+        pricing: {
+          inputPerM: 3,
+          outputPerM: 15,
+          cacheWritePerM: 3.75,
+          cacheReadPerM: 0.3,
+          currency: "USD",
+          unit: "per_million_tokens"
+        }
+      },
+      {
+        name: "claude-3-5-haiku@20241022",
+        displayName: "Claude 3.5 Haiku (Vertex)",
+        maxInputTokens: 2e5,
+        supportedFileTypes: ["pdf", "image"],
+        pricing: {
+          inputPerM: 0.8,
+          outputPerM: 4,
+          cacheWritePerM: 1,
+          cacheReadPerM: 0.08,
+          currency: "USD",
+          unit: "per_million_tokens"
+        }
+      }
+    ],
+    baseURLSupport: "none",
+    // Auto-constructed from projectId and region
+    supportedFileTypes: ["pdf", "image", "audio"]
+  },
+  // Amazon Bedrock - AWS-hosted gateway for Claude, Nova, and more
+  // Auth: AWS credentials (env vars) or Bedrock API key (AWS_BEARER_TOKEN_BEDROCK)
+  //
+  // Cross-region inference: Auto-added for anthropic.* and amazon.* models
+  // supportsCustomModels: true allows users to add custom model IDs beyond the fixed list
+  bedrock: {
+    supportsCustomModels: true,
+    models: [
+      // Claude 4.5 models (latest)
+      {
+        name: "anthropic.claude-sonnet-4-5-20250929-v1:0",
+        displayName: "Claude 4.5 Sonnet",
+        maxInputTokens: 2e5,
+        default: true,
+        supportedFileTypes: ["pdf", "image"],
+        pricing: {
+          inputPerM: 3,
+          outputPerM: 15,
+          cacheWritePerM: 3.75,
+          cacheReadPerM: 0.3,
+          currency: "USD",
+          unit: "per_million_tokens"
+        }
+      },
+      {
+        name: "anthropic.claude-haiku-4-5-20251001-v1:0",
+        displayName: "Claude 4.5 Haiku",
+        maxInputTokens: 2e5,
+        supportedFileTypes: ["pdf", "image"],
+        pricing: {
+          inputPerM: 1,
+          outputPerM: 5,
+          cacheWritePerM: 1.25,
+          cacheReadPerM: 0.1,
+          currency: "USD",
+          unit: "per_million_tokens"
+        }
+      },
+      {
+        name: "anthropic.claude-opus-4-5-20251101-v1:0",
+        displayName: "Claude 4.5 Opus",
+        maxInputTokens: 2e5,
+        supportedFileTypes: ["pdf", "image"],
+        pricing: {
+          inputPerM: 5,
+          outputPerM: 25,
+          cacheWritePerM: 6.25,
+          cacheReadPerM: 0.5,
+          currency: "USD",
+          unit: "per_million_tokens"
+        }
+      },
+      // Amazon Nova models
+      {
+        name: "amazon.nova-premier-v1:0",
+        displayName: "Nova Premier",
+        maxInputTokens: 1e6,
+        supportedFileTypes: ["image"],
+        pricing: {
+          inputPerM: 2.5,
+          outputPerM: 12.5,
+          currency: "USD",
+          unit: "per_million_tokens"
+        }
+      },
+      {
+        name: "amazon.nova-pro-v1:0",
+        displayName: "Nova Pro",
+        maxInputTokens: 3e5,
+        supportedFileTypes: ["pdf", "image"],
+        pricing: {
+          inputPerM: 0.8,
+          outputPerM: 3.2,
+          cacheReadPerM: 0.2,
+          currency: "USD",
+          unit: "per_million_tokens"
+        }
+      },
+      {
+        name: "amazon.nova-lite-v1:0",
+        displayName: "Nova Lite",
+        maxInputTokens: 3e5,
+        supportedFileTypes: ["pdf", "image"],
+        pricing: {
+          inputPerM: 0.06,
+          outputPerM: 0.24,
+          cacheReadPerM: 0.015,
+          currency: "USD",
+          unit: "per_million_tokens"
+        }
+      },
+      {
+        name: "amazon.nova-micro-v1:0",
+        displayName: "Nova Micro",
+        maxInputTokens: 128e3,
+        supportedFileTypes: [],
+        pricing: {
+          inputPerM: 0.035,
+          outputPerM: 0.14,
+          cacheReadPerM: 875e-5,
+          currency: "USD",
+          unit: "per_million_tokens"
+        }
+      },
+      // OpenAI GPT-OSS
+      {
+        name: "openai.gpt-oss-120b-1:0",
+        displayName: "GPT-OSS 120B",
+        maxInputTokens: 128e3,
+        supportedFileTypes: [],
+        pricing: {
+          inputPerM: 0.15,
+          outputPerM: 0.6,
+          currency: "USD",
+          unit: "per_million_tokens"
+        }
+      },
+      {
+        name: "openai.gpt-oss-20b-1:0",
+        displayName: "GPT-OSS 20B",
+        maxInputTokens: 128e3,
+        supportedFileTypes: [],
+        pricing: {
+          inputPerM: 0.07,
+          outputPerM: 0.3,
+          currency: "USD",
+          unit: "per_million_tokens"
+        }
+      },
+      // Qwen
+      {
+        name: "qwen.qwen3-coder-30b-a3b-v1:0",
+        displayName: "Qwen3 Coder 30B",
+        maxInputTokens: 262144,
+        supportedFileTypes: [],
+        pricing: {
+          inputPerM: 0.15,
+          outputPerM: 0.6,
+          currency: "USD",
+          unit: "per_million_tokens"
+        }
+      },
+      {
+        name: "qwen.qwen3-coder-480b-a35b-v1:0",
+        displayName: "Qwen3 Coder 480B",
+        maxInputTokens: 262144,
+        supportedFileTypes: [],
+        pricing: {
+          inputPerM: 0.22,
+          outputPerM: 1.8,
+          currency: "USD",
+          unit: "per_million_tokens"
+        }
+      }
+    ],
+    baseURLSupport: "none",
+    // Auto-constructed from region
+    supportedFileTypes: ["pdf", "image"]
+  },
+  // Native local model execution via node-llama-cpp
+  // Runs GGUF models directly on the machine using Metal/CUDA/Vulkan acceleration
+  // Models are downloaded from HuggingFace and stored in ~/.dexto/models/
+  local: {
+    models: [],
+    // Populated dynamically from local model registry
+    baseURLSupport: "none",
+    // No external server needed
+    supportedFileTypes: ["image"],
+    // Vision support depends on model capabilities
+    supportsCustomModels: true
+    // Allow any GGUF model path
+  },
+  // Ollama server integration
+  // Uses Ollama's OpenAI-compatible API for local model inference
+  // Requires Ollama to be installed and running (default: http://localhost:11434)
+  ollama: {
+    models: [],
+    // Populated dynamically from Ollama API
+    baseURLSupport: "optional",
+    // Default: http://localhost:11434, can be customized
+    supportedFileTypes: ["image"],
+    // Vision support depends on model
+    supportsCustomModels: true
+    // Accept any Ollama model name
   }
+  // TODO: Add 'dexto' provider (similar to openrouter, uses https://api.dexto.ai/v1)
 };
+function stripBedrockRegionPrefix(model) {
+  if (model.startsWith("eu.") || model.startsWith("us.")) {
+    return model.slice(3);
+  }
+  if (model.startsWith("global.")) {
+    return model.slice(7);
+  }
+  return model;
+}
 function getDefaultModelForProvider(provider) {
   const providerInfo = LLM_REGISTRY[provider];
   return providerInfo.models.find((m) => m.default)?.name || null;
@@ -822,7 +1300,8 @@ function getSupportedModels(provider) {
 }
 function getMaxInputTokensForModel(provider, model, logger) {
   const providerInfo = LLM_REGISTRY[provider];
-  const modelInfo = providerInfo.models.find((m) => m.name.toLowerCase() === model.toLowerCase());
+  const normalizedModel = stripBedrockRegionPrefix(model).toLowerCase();
+  const modelInfo = providerInfo.models.find((m) => m.name.toLowerCase() === normalizedModel);
   if (!modelInfo) {
     const supportedModels = getSupportedModels(provider).join(", ");
     logger?.error(
@@ -835,13 +1314,14 @@ function getMaxInputTokensForModel(provider, model, logger) {
 }
 function isValidProviderModel(provider, model) {
   const providerInfo = LLM_REGISTRY[provider];
-  return providerInfo.models.some((m) => m.name.toLowerCase() === model.toLowerCase());
+  const normalizedModel = stripBedrockRegionPrefix(model).toLowerCase();
+  return providerInfo.models.some((m) => m.name.toLowerCase() === normalizedModel);
 }
 function getProviderFromModel(model) {
-  const lowerModel = model.toLowerCase();
+  const normalizedModel = stripBedrockRegionPrefix(model).toLowerCase();
   for (const provider of import_types.LLM_PROVIDERS) {
     const info = LLM_REGISTRY[provider];
-    if (info.models.some((m) => m.name.toLowerCase() === lowerModel)) {
+    if (info.models.some((m) => m.name.toLowerCase() === normalizedModel)) {
       return provider;
     }
   }
@@ -862,12 +1342,34 @@ function acceptsAnyModel(provider) {
   const providerInfo = LLM_REGISTRY[provider];
   return providerInfo.models.length === 0;
 }
+function supportsCustomModels(provider) {
+  const providerInfo = LLM_REGISTRY[provider];
+  return providerInfo.supportsCustomModels === true;
+}
+const API_KEY_OPTIONAL_PROVIDERS = /* @__PURE__ */ new Set([
+  "local",
+  // Native node-llama-cpp execution - no auth needed
+  "ollama",
+  // Ollama server - no auth needed by default
+  "openai-compatible",
+  // vLLM, LocalAI - often no auth needed
+  "litellm",
+  // Self-hosted proxy - handles auth internally
+  "vertex",
+  // Uses Google Cloud ADC (Application Default Credentials)
+  "bedrock"
+  // Uses AWS credentials (access key + secret or IAM role)
+]);
+function requiresApiKey(provider) {
+  return !API_KEY_OPTIONAL_PROVIDERS.has(provider);
+}
 function getSupportedFileTypesForModel(provider, model) {
   const providerInfo = LLM_REGISTRY[provider];
   if (acceptsAnyModel(provider)) {
     return providerInfo.supportedFileTypes;
   }
-  const modelInfo = providerInfo.models.find((m) => m.name.toLowerCase() === model.toLowerCase());
+  const normalizedModel = stripBedrockRegionPrefix(model).toLowerCase();
+  const modelInfo = providerInfo.models.find((m) => m.name.toLowerCase() === normalizedModel);
   if (!modelInfo) {
     throw import_errors.LLMError.unknownModel(provider, model);
   }
@@ -946,6 +1448,19 @@ function getEffectiveMaxInputTokens(config, logger) {
       }
     }
   }
+  if (config.provider === "openrouter") {
+    const contextLength = (0, import_openrouter_model_registry.getOpenRouterModelContextLength)(config.model);
+    if (contextLength !== null) {
+      logger.debug(
+        `Using maxInputTokens from OpenRouter registry for ${config.model}: ${contextLength}`
+      );
+      return contextLength;
+    }
+    logger.warn(
+      `OpenRouter model ${config.model} not found in cache, defaulting to ${DEFAULT_MAX_INPUT_TOKENS} tokens`
+    );
+    return DEFAULT_MAX_INPUT_TOKENS;
+  }
   if (config.baseURL) {
     logger.warn(
       `baseURL is set but maxInputTokens is missing. Defaulting to ${DEFAULT_MAX_INPUT_TOKENS}. Provide 'maxInputTokens' in configuration to avoid default fallback.`
@@ -970,6 +1485,12 @@ function getEffectiveMaxInputTokens(config, logger) {
     return registryMaxInputTokens;
   } catch (error) {
     if (error instanceof import_DextoRuntimeError.DextoRuntimeError && error.code === import_error_codes.LLMErrorCode.MODEL_UNKNOWN) {
+      if (supportsCustomModels(config.provider)) {
+        logger.debug(
+          `Custom model ${config.model} not in ${config.provider} registry, defaulting to ${DEFAULT_MAX_INPUT_TOKENS} tokens`
+        );
+        return DEFAULT_MAX_INPUT_TOKENS;
+      }
       logger.error(
         `Registry lookup failed for ${config.provider}/${config.model}: ${error.message}. Effective maxInputTokens cannot be determined.`
       );
@@ -985,7 +1506,8 @@ function getModelPricing(provider, model) {
   if (acceptsAnyModel(provider)) {
     return void 0;
   }
-  const modelInfo = providerInfo.models.find((m) => m.name.toLowerCase() === model.toLowerCase());
+  const normalizedModel = stripBedrockRegionPrefix(model).toLowerCase();
+  const modelInfo = providerInfo.models.find((m) => m.name.toLowerCase() === normalizedModel);
   return modelInfo?.pricing;
 }
 function getModelDisplayName(model, provider) {
@@ -999,7 +1521,8 @@ function getModelDisplayName(model, provider) {
   if (!providerInfo || acceptsAnyModel(resolvedProvider)) {
     return model;
   }
-  const modelInfo = providerInfo.models.find((m) => m.name.toLowerCase() === model.toLowerCase());
+  const normalizedModel = stripBedrockRegionPrefix(model).toLowerCase();
+  const modelInfo = providerInfo.models.find((m) => m.name.toLowerCase() === normalizedModel);
   return modelInfo?.displayName ?? model;
 }
 function calculateCost(usage, pricing) {
@@ -1030,7 +1553,10 @@ function calculateCost(usage, pricing) {
   getSupportedProviders,
   isValidProviderModel,
   modelSupportsFileType,
+  requiresApiKey,
   requiresBaseURL,
+  stripBedrockRegionPrefix,
   supportsBaseURL,
+  supportsCustomModels,
   validateModelFileSupport
 });