@prestyj/ai 5.13.0 → 5.16.1

package/dist/index.d.cts

@@ -3,6 +3,8 @@ import Anthropic from '@anthropic-ai/sdk';
 import OpenAI from 'openai';
 
 type Provider = "anthropic" | "xiaomi" | "openai" | "gemini" | "glm" | "moonshot" | "minimax" | "deepseek" | "openrouter" | "sakana" | "xai" | "palsu"
+/** Hugging Face Inference Providers router (OpenAI-compatible). */
+ | "huggingface"
 /** Locally hosted OpenAI-compatible server (Ollama, LM Studio, llama.cpp, vLLM). */
  | "local";
 type ThinkingLevel = "low" | "medium" | "high" | "xhigh" | "max" | "ultra";

package/dist/index.d.ts

@@ -3,6 +3,8 @@ import Anthropic from '@anthropic-ai/sdk';
 import OpenAI from 'openai';
 
 type Provider = "anthropic" | "xiaomi" | "openai" | "gemini" | "glm" | "moonshot" | "minimax" | "deepseek" | "openrouter" | "sakana" | "xai" | "palsu"
+/** Hugging Face Inference Providers router (OpenAI-compatible). */
+ | "huggingface"
 /** Locally hosted OpenAI-compatible server (Ollama, LM Studio, llama.cpp, vLLM). */
  | "local";
 type ThinkingLevel = "low" | "medium" | "high" | "xhigh" | "max" | "ultra";

package/dist/index.js

@@ -59,6 +59,7 @@ var PROVIDER_DISPLAY = {
   openrouter: "OpenRouter",
   sakana: "Sakana",
   xai: "xAI (Grok)",
+  huggingface: "Hugging Face",
   xiaomi: "Xiaomi (MiMo)",
   minimax: "MiniMax"
 };
@@ -126,7 +127,7 @@ function formatError(err) {
         provider: err.provider,
         statusCode: err.statusCode,
         ...err.requestId ? { requestId: err.requestId } : {},
-        guidance: "Request access via your Anthropic account team (see platform.claude.com/docs/en/about-claude/models/overview), or switch to Claude Fable 5 via the model selector \u2014 same underlying model, generally available."
+        guidance: "Request access via your Anthropic account team (see platform.claude.com/docs/en/about-claude/models/overview), or switch to Claude Fable 5.1 via the model selector \u2014 same underlying model, generally available."
       };
     }
     if (isUsageLimitError(err)) {
@@ -1554,7 +1555,15 @@ async function* runStream(options) {
           yield keepalive;
           break;
         }
-        // message_stop — loop exits naturally
+        // message_stop — loop exits naturally.
+        //
+        // Deliberately NOT breaking early here. Breaking makes the SDK iterator
+        // run `if (!done) controller.abort()` in its `finally`
+        // (core/streaming.js:97), which tears the connection down instead of
+        // returning it to the keep-alive pool — every turn would then pay a
+        // fresh TLS handshake. Draining to the end is what every other Anthropic
+        // client does, and the stall it guards against is handled by the agent
+        // loop's idle timeout.
         default:
           yield keepalive;
           break;
@@ -2964,6 +2973,7 @@ var CODE_ASSIST_SUPPORTED_MODELS = /* @__PURE__ */ new Set([
   "gemini-3.5-flash",
   "gemini-3-flash",
   "gemini-3.1-flash-lite",
+  "gemini-3.7-flash",
   "gemini-2.5-pro",
   "gemini-2.5-flash",
   "gemma-4-31b-it",
@@ -2986,13 +2996,14 @@ var ACCOUNT_GATED_MODELS = /* @__PURE__ */ new Set([
   "gemini-3-flash",
   "gemini-3.5-flash",
   "gemini-3.1-pro-preview",
-  "gemini-3.1-pro-preview-customtools"
+  "gemini-3.1-pro-preview-customtools",
+  "gemini-3.7-flash"
 ]);
 function accountGatedMessage(model) {
   return `Your Google account isn't entitled to "${model}" over Gemini Code Assist OAuth, so the API reports it as not found. This is an account-access limit, not a ezcoder bug.`;
 }
 function accountGatedHint() {
-  return `Newer Gemini models (3.5 Flash, 3.1 Pro Preview) are available only to Code Assist Standard/Enterprise accounts with preview/GA access enabled by a cloud admin \u2014 free/personal accounts usually can't call them. Switch to Gemini 3.1 Flash Lite (it works on this account) with /model, or sign in with a Code Assist Standard/Enterprise account that has preview access.`;
+  return `Newer Gemini models (3.7 Flash, 3.5 Flash, 3.1 Pro Preview) are available only to Code Assist Standard/Enterprise accounts with preview/GA access enabled by a cloud admin \u2014 free/personal accounts usually can't call them. Switch to Gemini 3.1 Flash Lite (it works on this account) with /model, or sign in with a Code Assist Standard/Enterprise account that has preview access.`;
 }
 function formatErrorMessage(status, body, model) {
   if (status === 404 && !CODE_ASSIST_SUPPORTED_MODELS.has(model)) {
@@ -3676,6 +3687,18 @@ providerRegistry.register("openrouter", {
     baseUrl: options.baseUrl ?? "https://openrouter.ai/api/v1"
   })
 });
+providerRegistry.register("huggingface", {
+  // Hugging Face Inference Providers router — one HF token (hf.co/settings/tokens,
+  // "Make calls to Inference Providers" permission) routes to whichever hosted
+  // backend serves each open model. Chat Completions-compatible; model ids are
+  // Hub repo paths ("Qwen/Qwen3-Coder-480B-A35B-Instruct"), optionally with an
+  // ":auto"/":fastest"/":cheapest" provider-selection suffix. Billing follows
+  // each backend's per-token rates on the HF account (small free tier).
+  stream: (options) => streamOpenAI({
+    ...options,
+    baseUrl: options.baseUrl ?? "https://router.huggingface.co/v1"
+  })
+});
 providerRegistry.register("sakana", {
   // Sakana Fugu is a multi-agent system exposed as a standard LLM through the
   // OpenAI-compatible Sakana API. We ride the Chat Completions transport (the