llm-strings 1.1.2 → 1.3.0

package/dist/index.cjs

@@ -20,50 +20,163 @@ var __toCommonJS = (mod) => __copyProps(__defProp({}, "__esModule", { value: tru
 // src/index.ts
 var index_exports = {};
 __export(index_exports, {
+  HOST_ALIASES: () => HOST_ALIASES,
   build: () => build,
   normalize: () => normalize,
   parse: () => parse,
+  resolveHostAlias: () => resolveHostAlias,
   validate: () => validate
 });
 module.exports = __toCommonJS(index_exports);
 
-// src/parse.ts
-function parse(connectionString) {
-  const url = new URL(connectionString);
-  if (url.protocol !== "llm:") {
-    throw new Error(
-      `Invalid scheme: expected "llm://", got "${url.protocol}//"`
-    );
+// src/provider-core.ts
+function hasOwn(object, key) {
+  return Object.prototype.hasOwnProperty.call(object, key);
+}
+var HOST_ALIASES = {
+  openai: "api.openai.com",
+  azure: "models.inference.ai.azure.com",
+  anthropic: "api.anthropic.com",
+  google: "generativelanguage.googleapis.com",
+  "google-vertex": "aiplatform.googleapis.com",
+  aistudio: "generativelanguage.googleapis.com",
+  mistral: "api.mistral.ai",
+  cohere: "api.cohere.com",
+  bedrock: "bedrock-runtime.us-east-1.amazonaws.com",
+  openrouter: "openrouter.ai",
+  vercel: "gateway.ai.vercel.app",
+  alibaba: "dashscope-intl.aliyuncs.com",
+  alibabacloud: "dashscope-intl.aliyuncs.com",
+  dashscope: "dashscope-intl.aliyuncs.com",
+  groq: "api.groq.com",
+  fal: "fal.run",
+  fireworks: "api.fireworks.ai",
+  fireworksai: "api.fireworks.ai",
+  "black-forest-labs": "api.bfl.ai",
+  bfl: "api.bfl.ai",
+  deepseek: "api.deepseek.com",
+  moonshotai: "api.moonshot.ai",
+  moonshot: "api.moonshot.ai",
+  perplexity: "api.perplexity.ai",
+  venice: "api.venice.ai",
+  parasail: "api.parasail.io",
+  deepinfra: "api.deepinfra.com",
+  atlascloud: "api.atlascloud.ai",
+  novita: "api.novita.ai",
+  novitaai: "api.novita.ai",
+  grok: "api.x.ai",
+  xai: "api.x.ai",
+  together: "api.together.xyz",
+  togetherai: "api.together.xyz",
+  cerebras: "api.cerebras.ai",
+  replicate: "api.replicate.com",
+  prodia: "api.prodia.com",
+  luma: "api.lumalabs.ai",
+  bytedance: "ark.cn-beijing.volces.com",
+  kling: "api.klingai.com",
+  elevenlabs: "api.elevenlabs.io",
+  assemblyai: "api.assemblyai.com",
+  deepgram: "api.deepgram.com",
+  gladia: "api.gladia.io",
+  lmnt: "api.lmnt.com",
+  hume: "api.hume.ai",
+  revai: "api.rev.ai",
+  baseten: "api.baseten.co",
+  huggingface: "api-inference.huggingface.co",
+  wandb: "api.inference.wandb.ai",
+  weightsandbiases: "api.inference.wandb.ai",
+  baidu: "qianfan.baidubce.com",
+  qianfan: "qianfan.baidubce.com",
+  vertex: "aiplatform.googleapis.com",
+  xiaomi: "api.xiaomimimo.com",
+  minimax: "api.minimax.io"
+};
+var HOST_ALIAS_PROVIDERS = {
+  aistudio: "google",
+  vertex: "google-vertex",
+  grok: "xai",
+  bfl: "black-forest-labs",
+  moonshot: "moonshotai",
+  alibaba: "alibaba",
+  alibabacloud: "alibaba",
+  dashscope: "alibaba",
+  togetherai: "together",
+  fireworksai: "fireworks"
+};
+function readProcessEnv() {
+  return typeof process !== "undefined" && process.env ? process.env : {};
+}
+function normalizeHostValue(value) {
+  const trimmed = value.trim();
+  if (!trimmed) return trimmed;
+  try {
+    if (trimmed.includes("://")) {
+      return new URL(trimmed).host;
+    }
+  } catch {
   }
-  const host = url.hostname;
-  const model = url.pathname.replace(/^\//, "");
-  const label = url.username || void 0;
-  const apiKey = url.password || void 0;
-  const params = {};
-  for (const [key, value] of url.searchParams) {
-    params[key] = value;
+  return trimmed.replace(/^\/\//, "").split("/")[0] ?? trimmed;
+}
+function envHostOverride(alias, env) {
+  const upper = alias.toUpperCase();
+  const override = env[`LLM_STRINGS_${upper}_HOST`] ?? env[`LLM_STRINGS_HOST_${upper}`];
+  return override?.trim() ? override : void 0;
+}
+function resolveHostAlias(host, env = readProcessEnv()) {
+  const normalizedHost = host.toLowerCase();
+  if (!hasOwn(HOST_ALIASES, normalizedHost)) {
+    return { host };
   }
+  const alias = normalizedHost;
+  const override = envHostOverride(alias, env);
   return {
-    raw: connectionString,
-    host,
-    model,
-    label,
-    apiKey,
-    params
+    host: normalizeHostValue(override ?? HOST_ALIASES[alias]),
+    alias
   };
 }
-function build(config) {
-  const auth = config.label || config.apiKey ? `${config.label ?? ""}${config.apiKey ? `:${config.apiKey}` : ""}@` : "";
-  const query = new URLSearchParams(config.params).toString();
-  const qs = query ? `?${query}` : "";
-  return `llm://${auth}${config.host}/${config.model}${qs}`;
+function providerFromHostAlias(alias) {
+  const normalizedAlias = alias.toLowerCase();
+  if (hasOwn(PROVIDER_PARAMS, normalizedAlias)) {
+    return normalizedAlias;
+  }
+  if (hasOwn(HOST_ALIAS_PROVIDERS, normalizedAlias)) {
+    return HOST_ALIAS_PROVIDERS[normalizedAlias];
+  }
+  return void 0;
 }
-
-// src/provider-core.ts
 function detectProvider(host) {
+  host = host.toLowerCase();
   if (host.includes("openrouter")) return "openrouter";
   if (host.includes("gateway.ai.vercel")) return "vercel";
   if (host.includes("amazonaws") || host.includes("bedrock")) return "bedrock";
+  if (host.includes("aiplatform.googleapis")) return "google-vertex";
+  if (host.includes("api.x.ai")) return "xai";
+  if (host.includes("groq")) return "groq";
+  if (host.includes("fal.run") || host.includes("fal.ai")) return "fal";
+  if (host.includes("deepinfra")) return "deepinfra";
+  if (host.includes("bfl.ai")) return "black-forest-labs";
+  if (host.includes("together")) return "together";
+  if (host.includes("fireworks")) return "fireworks";
+  if (host.includes("deepseek")) return "deepseek";
+  if (host.includes("moonshot")) return "moonshotai";
+  if (host.includes("perplexity")) return "perplexity";
+  if (host.includes("dashscope") || host.includes("aliyuncs")) return "alibaba";
+  if (host.includes("cerebras")) return "cerebras";
+  if (host.includes("replicate")) return "replicate";
+  if (host.includes("prodia")) return "prodia";
+  if (host.includes("lumalabs") || host.includes("luma")) return "luma";
+  if (host.includes("volces") || host.includes("bytedance")) return "bytedance";
+  if (host.includes("kling")) return "kling";
+  if (host.includes("elevenlabs")) return "elevenlabs";
+  if (host.includes("assemblyai")) return "assemblyai";
+  if (host.includes("deepgram")) return "deepgram";
+  if (host.includes("gladia")) return "gladia";
+  if (host.includes("lmnt")) return "lmnt";
+  if (host.includes("hume")) return "hume";
+  if (host.includes("rev.ai")) return "revai";
+  if (host.includes("baseten")) return "baseten";
+  if (host.includes("huggingface")) return "huggingface";
+  if (host.includes("azure")) return "azure";
   if (host.includes("openai")) return "openai";
   if (host.includes("anthropic") || host.includes("claude")) return "anthropic";
   if (host.includes("googleapis") || host.includes("google")) return "google";
@@ -118,6 +231,33 @@ var ALIASES = {
   cachePoint: "cache",
   cache_point: "cache"
 };
+var OPENAI_COMPATIBLE_PARAMS = {
+  temperature: "temperature",
+  max_tokens: "max_tokens",
+  top_p: "top_p",
+  top_k: "top_k",
+  frequency_penalty: "frequency_penalty",
+  presence_penalty: "presence_penalty",
+  stop: "stop",
+  n: "n",
+  seed: "seed",
+  stream: "stream",
+  effort: "reasoning_effort"
+};
+var GOOGLE_COMPATIBLE_PARAMS = {
+  temperature: "temperature",
+  max_tokens: "maxOutputTokens",
+  top_p: "topP",
+  top_k: "topK",
+  frequency_penalty: "frequencyPenalty",
+  presence_penalty: "presencePenalty",
+  stop: "stopSequences",
+  n: "candidateCount",
+  stream: "stream",
+  seed: "seed",
+  responseMimeType: "responseMimeType",
+  responseSchema: "responseSchema"
+};
 var PROVIDER_PARAMS = {
   openai: {
     temperature: "temperature",
@@ -131,6 +271,7 @@ var PROVIDER_PARAMS = {
     stream: "stream",
     effort: "reasoning_effort"
   },
+  azure: OPENAI_COMPATIBLE_PARAMS,
   anthropic: {
     temperature: "temperature",
     max_tokens: "max_tokens",
@@ -156,6 +297,7 @@ var PROVIDER_PARAMS = {
     responseMimeType: "responseMimeType",
     responseSchema: "responseSchema"
   },
+  "google-vertex": GOOGLE_COMPATIBLE_PARAMS,
   mistral: {
     temperature: "temperature",
     max_tokens: "max_tokens",
@@ -219,15 +361,175 @@ var PROVIDER_PARAMS = {
     seed: "seed",
     stream: "stream",
     effort: "reasoning_effort"
+  },
+  xai: OPENAI_COMPATIBLE_PARAMS,
+  groq: OPENAI_COMPATIBLE_PARAMS,
+  fal: {},
+  deepinfra: OPENAI_COMPATIBLE_PARAMS,
+  "black-forest-labs": {},
+  together: OPENAI_COMPATIBLE_PARAMS,
+  fireworks: OPENAI_COMPATIBLE_PARAMS,
+  deepseek: OPENAI_COMPATIBLE_PARAMS,
+  moonshotai: OPENAI_COMPATIBLE_PARAMS,
+  perplexity: OPENAI_COMPATIBLE_PARAMS,
+  alibaba: OPENAI_COMPATIBLE_PARAMS,
+  cerebras: OPENAI_COMPATIBLE_PARAMS,
+  replicate: {},
+  prodia: {},
+  luma: {},
+  bytedance: {},
+  kling: {},
+  elevenlabs: {},
+  assemblyai: {},
+  deepgram: {},
+  gladia: {},
+  lmnt: {},
+  hume: {},
+  revai: {},
+  baseten: OPENAI_COMPATIBLE_PARAMS,
+  huggingface: OPENAI_COMPATIBLE_PARAMS
+};
+var OPENAI_COMPATIBLE_PARAM_SPECS = {
+  temperature: {
+    type: "number",
+    min: 0,
+    max: 2,
+    default: 0.7,
+    description: "Controls randomness"
+  },
+  max_tokens: {
+    type: "number",
+    min: 1,
+    default: 4096,
+    description: "Maximum output tokens"
+  },
+  top_p: {
+    type: "number",
+    min: 0,
+    max: 1,
+    default: 1,
+    description: "Nucleus sampling"
+  },
+  top_k: {
+    type: "number",
+    min: 0,
+    default: 40,
+    description: "Top-K sampling"
+  },
+  frequency_penalty: {
+    type: "number",
+    min: -2,
+    max: 2,
+    default: 0,
+    description: "Penalize frequent tokens"
+  },
+  presence_penalty: {
+    type: "number",
+    min: -2,
+    max: 2,
+    default: 0,
+    description: "Penalize repeated topics"
+  },
+  stop: { type: "string", description: "Stop sequences" },
+  n: { type: "number", min: 1, default: 1, description: "Completions count" },
+  seed: { type: "number", description: "Random seed" },
+  stream: { type: "boolean", default: false, description: "Stream response" },
+  reasoning_effort: {
+    type: "string",
+    values: ["none", "minimal", "low", "medium", "high", "xhigh"],
+    default: "medium",
+    description: "Reasoning effort"
   }
 };
+var GOOGLE_COMPATIBLE_PARAM_SPECS = {
+  temperature: {
+    type: "number",
+    min: 0,
+    max: 2,
+    default: 0.7,
+    description: "Controls randomness"
+  },
+  maxOutputTokens: {
+    type: "number",
+    min: 1,
+    default: 4096,
+    description: "Maximum output tokens"
+  },
+  topP: {
+    type: "number",
+    min: 0,
+    max: 1,
+    default: 1,
+    description: "Nucleus sampling"
+  },
+  topK: {
+    type: "number",
+    min: 0,
+    default: 40,
+    description: "Top-K sampling"
+  },
+  frequencyPenalty: {
+    type: "number",
+    min: -2,
+    max: 2,
+    default: 0,
+    description: "Penalize frequent tokens"
+  },
+  presencePenalty: {
+    type: "number",
+    min: -2,
+    max: 2,
+    default: 0,
+    description: "Penalize repeated topics"
+  },
+  stopSequences: { type: "string", description: "Stop sequences" },
+  candidateCount: {
+    type: "number",
+    min: 1,
+    default: 1,
+    description: "Candidate count"
+  },
+  stream: { type: "boolean", default: false, description: "Stream response" },
+  seed: { type: "number", description: "Random seed" },
+  responseMimeType: { type: "string", description: "Response MIME type" },
+  responseSchema: { type: "string", description: "Response schema" }
+};
 var PARAM_SPECS = {
   openai: {
-    temperature: { type: "number", min: 0, max: 2, default: 0.7, description: "Controls randomness" },
-    max_tokens: { type: "number", min: 1, default: 4096, description: "Maximum output tokens" },
-    top_p: { type: "number", min: 0, max: 1, default: 1, description: "Nucleus sampling" },
-    frequency_penalty: { type: "number", min: -2, max: 2, default: 0, description: "Penalize frequent tokens" },
-    presence_penalty: { type: "number", min: -2, max: 2, default: 0, description: "Penalize repeated topics" },
+    temperature: {
+      type: "number",
+      min: 0,
+      max: 2,
+      default: 0.7,
+      description: "Controls randomness"
+    },
+    max_tokens: {
+      type: "number",
+      min: 1,
+      default: 4096,
+      description: "Maximum output tokens"
+    },
+    top_p: {
+      type: "number",
+      min: 0,
+      max: 1,
+      default: 1,
+      description: "Nucleus sampling"
+    },
+    frequency_penalty: {
+      type: "number",
+      min: -2,
+      max: 2,
+      default: 0,
+      description: "Penalize frequent tokens"
+    },
+    presence_penalty: {
+      type: "number",
+      min: -2,
+      max: 2,
+      default: 0,
+      description: "Penalize repeated topics"
+    },
     stop: { type: "string", description: "Stop sequences" },
     n: { type: "number", min: 1, default: 1, description: "Completions count" },
     seed: { type: "number", description: "Random seed" },
@@ -239,74 +541,291 @@ var PARAM_SPECS = {
       description: "Reasoning effort"
     }
   },
+  azure: OPENAI_COMPATIBLE_PARAM_SPECS,
   anthropic: {
-    temperature: { type: "number", min: 0, max: 1, default: 0.7, description: "Controls randomness" },
-    max_tokens: { type: "number", min: 1, default: 4096, description: "Maximum output tokens" },
-    top_p: { type: "number", min: 0, max: 1, default: 1, description: "Nucleus sampling" },
-    top_k: { type: "number", min: 0, default: 40, description: "Top-K sampling" },
+    temperature: {
+      type: "number",
+      min: 0,
+      max: 1,
+      default: 0.7,
+      description: "Controls randomness"
+    },
+    max_tokens: {
+      type: "number",
+      min: 1,
+      default: 4096,
+      description: "Maximum output tokens"
+    },
+    top_p: {
+      type: "number",
+      min: 0,
+      max: 1,
+      default: 1,
+      description: "Nucleus sampling"
+    },
+    top_k: {
+      type: "number",
+      min: 0,
+      default: 40,
+      description: "Top-K sampling"
+    },
     stop_sequences: { type: "string", description: "Stop sequences" },
     stream: { type: "boolean", default: false, description: "Stream response" },
-    effort: { type: "string", values: ["low", "medium", "high", "max"], default: "medium", description: "Thinking effort" },
-    cache_control: { type: "string", values: ["ephemeral"], default: "ephemeral", description: "Cache control" },
-    cache_ttl: { type: "string", values: ["5m", "1h"], default: "5m", description: "Cache TTL" }
+    effort: {
+      type: "string",
+      values: ["low", "medium", "high", "max"],
+      default: "medium",
+      description: "Thinking effort"
+    },
+    cache_control: {
+      type: "string",
+      values: ["ephemeral"],
+      default: "ephemeral",
+      description: "Cache control"
+    },
+    cache_ttl: {
+      type: "string",
+      values: ["5m", "1h"],
+      default: "5m",
+      description: "Cache TTL"
+    }
   },
   google: {
-    temperature: { type: "number", min: 0, max: 2, default: 0.7, description: "Controls randomness" },
-    maxOutputTokens: { type: "number", min: 1, default: 4096, description: "Maximum output tokens" },
-    topP: { type: "number", min: 0, max: 1, default: 1, description: "Nucleus sampling" },
-    topK: { type: "number", min: 0, default: 40, description: "Top-K sampling" },
-    frequencyPenalty: { type: "number", min: -2, max: 2, default: 0, description: "Penalize frequent tokens" },
-    presencePenalty: { type: "number", min: -2, max: 2, default: 0, description: "Penalize repeated topics" },
+    temperature: {
+      type: "number",
+      min: 0,
+      max: 2,
+      default: 0.7,
+      description: "Controls randomness"
+    },
+    maxOutputTokens: {
+      type: "number",
+      min: 1,
+      default: 4096,
+      description: "Maximum output tokens"
+    },
+    topP: {
+      type: "number",
+      min: 0,
+      max: 1,
+      default: 1,
+      description: "Nucleus sampling"
+    },
+    topK: {
+      type: "number",
+      min: 0,
+      default: 40,
+      description: "Top-K sampling"
+    },
+    frequencyPenalty: {
+      type: "number",
+      min: -2,
+      max: 2,
+      default: 0,
+      description: "Penalize frequent tokens"
+    },
+    presencePenalty: {
+      type: "number",
+      min: -2,
+      max: 2,
+      default: 0,
+      description: "Penalize repeated topics"
+    },
     stopSequences: { type: "string", description: "Stop sequences" },
-    candidateCount: { type: "number", min: 1, default: 1, description: "Candidate count" },
+    candidateCount: {
+      type: "number",
+      min: 1,
+      default: 1,
+      description: "Candidate count"
+    },
     stream: { type: "boolean", default: false, description: "Stream response" },
     seed: { type: "number", description: "Random seed" },
     responseMimeType: { type: "string", description: "Response MIME type" },
     responseSchema: { type: "string", description: "Response schema" }
   },
+  "google-vertex": GOOGLE_COMPATIBLE_PARAM_SPECS,
   mistral: {
-    temperature: { type: "number", min: 0, max: 1, default: 0.7, description: "Controls randomness" },
-    max_tokens: { type: "number", min: 1, default: 4096, description: "Maximum output tokens" },
-    top_p: { type: "number", min: 0, max: 1, default: 1, description: "Nucleus sampling" },
-    frequency_penalty: { type: "number", min: -2, max: 2, default: 0, description: "Penalize frequent tokens" },
-    presence_penalty: { type: "number", min: -2, max: 2, default: 0, description: "Penalize repeated topics" },
+    temperature: {
+      type: "number",
+      min: 0,
+      max: 1,
+      default: 0.7,
+      description: "Controls randomness"
+    },
+    max_tokens: {
+      type: "number",
+      min: 1,
+      default: 4096,
+      description: "Maximum output tokens"
+    },
+    top_p: {
+      type: "number",
+      min: 0,
+      max: 1,
+      default: 1,
+      description: "Nucleus sampling"
+    },
+    frequency_penalty: {
+      type: "number",
+      min: -2,
+      max: 2,
+      default: 0,
+      description: "Penalize frequent tokens"
+    },
+    presence_penalty: {
+      type: "number",
+      min: -2,
+      max: 2,
+      default: 0,
+      description: "Penalize repeated topics"
+    },
     stop: { type: "string", description: "Stop sequences" },
     n: { type: "number", min: 1, default: 1, description: "Completions count" },
     random_seed: { type: "number", description: "Random seed" },
     stream: { type: "boolean", default: false, description: "Stream response" },
-    safe_prompt: { type: "boolean", default: false, description: "Enable safe prompt" },
-    min_tokens: { type: "number", min: 0, default: 0, description: "Minimum tokens" }
+    safe_prompt: {
+      type: "boolean",
+      default: false,
+      description: "Enable safe prompt"
+    },
+    min_tokens: {
+      type: "number",
+      min: 0,
+      default: 0,
+      description: "Minimum tokens"
+    }
   },
   cohere: {
-    temperature: { type: "number", min: 0, max: 1, default: 0.7, description: "Controls randomness" },
-    max_tokens: { type: "number", min: 1, default: 4096, description: "Maximum output tokens" },
-    p: { type: "number", min: 0, max: 1, default: 1, description: "Nucleus sampling (p)" },
-    k: { type: "number", min: 0, max: 500, default: 40, description: "Top-K sampling (k)" },
-    frequency_penalty: { type: "number", min: 0, max: 1, default: 0, description: "Penalize frequent tokens" },
-    presence_penalty: { type: "number", min: 0, max: 1, default: 0, description: "Penalize repeated topics" },
+    temperature: {
+      type: "number",
+      min: 0,
+      max: 1,
+      default: 0.7,
+      description: "Controls randomness"
+    },
+    max_tokens: {
+      type: "number",
+      min: 1,
+      default: 4096,
+      description: "Maximum output tokens"
+    },
+    p: {
+      type: "number",
+      min: 0,
+      max: 1,
+      default: 1,
+      description: "Nucleus sampling (p)"
+    },
+    k: {
+      type: "number",
+      min: 0,
+      max: 500,
+      default: 40,
+      description: "Top-K sampling (k)"
+    },
+    frequency_penalty: {
+      type: "number",
+      min: 0,
+      max: 1,
+      default: 0,
+      description: "Penalize frequent tokens"
+    },
+    presence_penalty: {
+      type: "number",
+      min: 0,
+      max: 1,
+      default: 0,
+      description: "Penalize repeated topics"
+    },
     stop_sequences: { type: "string", description: "Stop sequences" },
     stream: { type: "boolean", default: false, description: "Stream response" },
     seed: { type: "number", description: "Random seed" }
   },
   bedrock: {
     // Converse API inferenceConfig params
-    temperature: { type: "number", min: 0, max: 1, default: 0.7, description: "Controls randomness" },
-    maxTokens: { type: "number", min: 1, default: 4096, description: "Maximum output tokens" },
-    topP: { type: "number", min: 0, max: 1, default: 1, description: "Nucleus sampling" },
-    topK: { type: "number", min: 0, default: 40, description: "Top-K sampling" },
+    temperature: {
+      type: "number",
+      min: 0,
+      max: 1,
+      default: 0.7,
+      description: "Controls randomness"
+    },
+    maxTokens: {
+      type: "number",
+      min: 1,
+      default: 4096,
+      description: "Maximum output tokens"
+    },
+    topP: {
+      type: "number",
+      min: 0,
+      max: 1,
+      default: 1,
+      description: "Nucleus sampling"
+    },
+    topK: {
+      type: "number",
+      min: 0,
+      default: 40,
+      description: "Top-K sampling"
+    },
     stopSequences: { type: "string", description: "Stop sequences" },
     stream: { type: "boolean", default: false, description: "Stream response" },
-    cache_control: { type: "string", values: ["ephemeral"], default: "ephemeral", description: "Cache control" },
-    cache_ttl: { type: "string", values: ["5m", "1h"], default: "5m", description: "Cache TTL" }
+    cache_control: {
+      type: "string",
+      values: ["ephemeral"],
+      default: "ephemeral",
+      description: "Cache control"
+    },
+    cache_ttl: {
+      type: "string",
+      values: ["5m", "1h"],
+      default: "5m",
+      description: "Cache TTL"
+    }
   },
   openrouter: {
     // Loose validation — proxies to many providers with varying ranges
-    temperature: { type: "number", min: 0, max: 2, default: 0.7, description: "Controls randomness" },
-    max_tokens: { type: "number", min: 1, default: 4096, description: "Maximum output tokens" },
-    top_p: { type: "number", min: 0, max: 1, default: 1, description: "Nucleus sampling" },
-    top_k: { type: "number", min: 0, default: 40, description: "Top-K sampling" },
-    frequency_penalty: { type: "number", min: -2, max: 2, default: 0, description: "Penalize frequent tokens" },
-    presence_penalty: { type: "number", min: -2, max: 2, default: 0, description: "Penalize repeated topics" },
+    temperature: {
+      type: "number",
+      min: 0,
+      max: 2,
+      default: 0.7,
+      description: "Controls randomness"
+    },
+    max_tokens: {
+      type: "number",
+      min: 1,
+      default: 4096,
+      description: "Maximum output tokens"
+    },
+    top_p: {
+      type: "number",
+      min: 0,
+      max: 1,
+      default: 1,
+      description: "Nucleus sampling"
+    },
+    top_k: {
+      type: "number",
+      min: 0,
+      default: 40,
+      description: "Top-K sampling"
+    },
+    frequency_penalty: {
+      type: "number",
+      min: -2,
+      max: 2,
+      default: 0,
+      description: "Penalize frequent tokens"
+    },
+    presence_penalty: {
+      type: "number",
+      min: -2,
+      max: 2,
+      default: 0,
+      description: "Penalize repeated topics"
+    },
     stop: { type: "string", description: "Stop sequences" },
     n: { type: "number", min: 1, default: 1, description: "Completions count" },
     seed: { type: "number", description: "Random seed" },
@@ -320,12 +839,46 @@ var PARAM_SPECS = {
   },
   vercel: {
     // Loose validation — proxies to many providers with varying ranges
-    temperature: { type: "number", min: 0, max: 2, default: 0.7, description: "Controls randomness" },
-    max_tokens: { type: "number", min: 1, default: 4096, description: "Maximum output tokens" },
-    top_p: { type: "number", min: 0, max: 1, default: 1, description: "Nucleus sampling" },
-    top_k: { type: "number", min: 0, default: 40, description: "Top-K sampling" },
-    frequency_penalty: { type: "number", min: -2, max: 2, default: 0, description: "Penalize frequent tokens" },
-    presence_penalty: { type: "number", min: -2, max: 2, default: 0, description: "Penalize repeated topics" },
+    temperature: {
+      type: "number",
+      min: 0,
+      max: 2,
+      default: 0.7,
+      description: "Controls randomness"
+    },
+    max_tokens: {
+      type: "number",
+      min: 1,
+      default: 4096,
+      description: "Maximum output tokens"
+    },
+    top_p: {
+      type: "number",
+      min: 0,
+      max: 1,
+      default: 1,
+      description: "Nucleus sampling"
+    },
+    top_k: {
+      type: "number",
+      min: 0,
+      default: 40,
+      description: "Top-K sampling"
+    },
+    frequency_penalty: {
+      type: "number",
+      min: -2,
+      max: 2,
+      default: 0,
+      description: "Penalize frequent tokens"
+    },
+    presence_penalty: {
+      type: "number",
+      min: -2,
+      max: 2,
+      default: 0,
+      description: "Penalize repeated topics"
+    },
     stop: { type: "string", description: "Stop sequences" },
     n: { type: "number", min: 1, default: 1, description: "Completions count" },
     seed: { type: "number", description: "Random seed" },
@@ -336,7 +889,33 @@ var PARAM_SPECS = {
       default: "medium",
       description: "Reasoning effort"
     }
-  }
+  },
+  xai: OPENAI_COMPATIBLE_PARAM_SPECS,
+  groq: OPENAI_COMPATIBLE_PARAM_SPECS,
+  fal: {},
+  deepinfra: OPENAI_COMPATIBLE_PARAM_SPECS,
+  "black-forest-labs": {},
+  together: OPENAI_COMPATIBLE_PARAM_SPECS,
+  fireworks: OPENAI_COMPATIBLE_PARAM_SPECS,
+  deepseek: OPENAI_COMPATIBLE_PARAM_SPECS,
+  moonshotai: OPENAI_COMPATIBLE_PARAM_SPECS,
+  perplexity: OPENAI_COMPATIBLE_PARAM_SPECS,
+  alibaba: OPENAI_COMPATIBLE_PARAM_SPECS,
+  cerebras: OPENAI_COMPATIBLE_PARAM_SPECS,
+  replicate: {},
+  prodia: {},
+  luma: {},
+  bytedance: {},
+  kling: {},
+  elevenlabs: {},
+  assemblyai: {},
+  deepgram: {},
+  gladia: {},
+  lmnt: {},
+  hume: {},
+  revai: {},
+  baseten: OPENAI_COMPATIBLE_PARAM_SPECS,
+  huggingface: OPENAI_COMPATIBLE_PARAM_SPECS
 };
 function isReasoningModel(model) {
   const name = model.includes("/") ? model.split("/").pop() : model;
@@ -352,7 +931,13 @@ function detectGatewaySubProvider(model) {
   const slash = model.indexOf("/");
   if (slash < 1) return void 0;
   const prefix = model.slice(0, slash);
-  const direct = ["openai", "anthropic", "google", "mistral", "cohere"];
+  const direct = [
+    "openai",
+    "anthropic",
+    "google",
+    "mistral",
+    "cohere"
+  ];
   return direct.find((p) => p === prefix);
 }
 var REASONING_MODEL_UNSUPPORTED = /* @__PURE__ */ new Set([
@@ -387,34 +972,124 @@ function bedrockSupportsCaching(model) {
 var CACHE_VALUES = {
   openai: void 0,
   // OpenAI auto-caches; no explicit param
+  azure: void 0,
   anthropic: "ephemeral",
   google: void 0,
   // Google uses explicit caching API, not a param
+  "google-vertex": void 0,
   mistral: void 0,
   cohere: void 0,
   bedrock: "ephemeral",
   // Supported for Claude models on Bedrock
   openrouter: void 0,
   // Depends on underlying provider
-  vercel: void 0
+  vercel: void 0,
   // Depends on underlying provider
+  xai: void 0,
+  groq: void 0,
+  fal: void 0,
+  deepinfra: void 0,
+  "black-forest-labs": void 0,
+  together: void 0,
+  fireworks: void 0,
+  deepseek: void 0,
+  moonshotai: void 0,
+  perplexity: void 0,
+  alibaba: void 0,
+  cerebras: void 0,
+  replicate: void 0,
+  prodia: void 0,
+  luma: void 0,
+  bytedance: void 0,
+  kling: void 0,
+  elevenlabs: void 0,
+  assemblyai: void 0,
+  deepgram: void 0,
+  gladia: void 0,
+  lmnt: void 0,
+  hume: void 0,
+  revai: void 0,
+  baseten: void 0,
+  huggingface: void 0
 };
 var CACHE_TTLS = {
   openai: void 0,
+  azure: void 0,
   anthropic: ["5m", "1h"],
   google: void 0,
+  "google-vertex": void 0,
   mistral: void 0,
   cohere: void 0,
   bedrock: ["5m", "1h"],
   // Claude on Bedrock uses same TTLs as direct Anthropic
   openrouter: void 0,
-  vercel: void 0
+  vercel: void 0,
+  xai: void 0,
+  groq: void 0,
+  fal: void 0,
+  deepinfra: void 0,
+  "black-forest-labs": void 0,
+  together: void 0,
+  fireworks: void 0,
+  deepseek: void 0,
+  moonshotai: void 0,
+  perplexity: void 0,
+  alibaba: void 0,
+  cerebras: void 0,
+  replicate: void 0,
+  prodia: void 0,
+  luma: void 0,
+  bytedance: void 0,
+  kling: void 0,
+  elevenlabs: void 0,
+  assemblyai: void 0,
+  deepgram: void 0,
+  gladia: void 0,
+  lmnt: void 0,
+  hume: void 0,
+  revai: void 0,
+  baseten: void 0,
+  huggingface: void 0
 };
 var DURATION_RE = /^\d+[mh]$/;
 
+// src/parse.ts
+function parse(connectionString) {
+  const url = new URL(connectionString);
+  if (url.protocol !== "llm:") {
+    throw new Error(
+      `Invalid scheme: expected "llm://", got "${url.protocol}//"`
+    );
+  }
+  const { host, alias: hostAlias } = resolveHostAlias(url.host);
+  const model = url.pathname.replace(/^\//, "");
+  const label = url.username || void 0;
+  const apiKey = url.password || void 0;
+  const params = {};
+  for (const [key, value] of url.searchParams) {
+    params[key] = value;
+  }
+  return {
+    raw: connectionString,
+    host,
+    hostAlias,
+    model,
+    label,
+    apiKey,
+    params
+  };
+}
+function build(config) {
+  const { host } = resolveHostAlias(config.host);
+  const auth = config.label || config.apiKey ? `${config.label ?? ""}${config.apiKey ? `:${config.apiKey}` : ""}@` : "";
+  const query = new URLSearchParams(config.params).toString();
+  const qs = query ? `?${query}` : "";
+  return `llm://${auth}${host}/${config.model}${qs}`;
+}
+
 // src/normalize.ts
 function normalize(config, options = {}) {
-  const provider = detectProvider(config.host);
+  const provider = (config.hostAlias ? providerFromHostAlias(config.hostAlias) : void 0) ?? detectProvider(config.host);
   const subProvider = provider && isGatewayProvider(provider) ? detectGatewaySubProvider(config.model) : void 0;
   const changes = [];
   const params = {};
@@ -599,11 +1274,7 @@ function validate(connectionString, options = {}) {
     }
     let spec = specs[key];
     if (subProvider && gatewayReverseMap && !spec) {
-      const result = lookupSubProviderSpec(
-        key,
-        gatewayReverseMap,
-        subProvider
-      );
+      const result = lookupSubProviderSpec(key, gatewayReverseMap, subProvider);
       spec = result.spec;
     }
     if (!spec) continue;
@@ -671,9 +1342,11 @@ function validate(connectionString, options = {}) {
 }
 // Annotate the CommonJS export names for ESM import in node:
 0 && (module.exports = {
+  HOST_ALIASES,
   build,
   normalize,
   parse,
+  resolveHostAlias,
   validate
 });
 //# sourceMappingURL=index.cjs.map