@jambonz/llm 0.4.2 → 0.4.4

package/dist/index.js

@@ -2331,21 +2331,21 @@ var azureOpenAIManifest = {
           label: "API Key",
           type: "password",
           required: true,
-          help: "Azure OpenAI resource key. Find it in the Azure portal under your OpenAI resource \u2192 Keys and Endpoint."
+          help: "In Azure AI Foundry, open the deployment and copy the Key from the Endpoint panel (eye icon to reveal). Same key is shown in the Azure portal under your OpenAI resource \u2192 Keys and Endpoint."
         },
         {
           name: "endpoint",
           label: "Endpoint",
           type: "url",
           required: true,
-          help: "Resource endpoint URL, e.g. https://my-resource.openai.azure.com. Shown in the Azure portal alongside the keys."
+          help: "Resource hostname only \u2014 no path, no query string. Example: https://my-resource.openai.azure.com. The Foundry deployment page shows a Target URI like https://my-resource.openai.azure.com/openai/responses?api-version=\u2026 Strip everything from /openai/ onward and paste only the part before it."
         },
         {
           name: "deployment",
           label: "Deployment Name",
           type: "text",
           required: true,
-          help: "The name you chose for the deployment in Azure AI Studio (NOT the underlying model id \u2014 deployments can be named anything)."
+          help: "The Name shown at the top of the deployment page in Azure AI Foundry (also under Deployment info \u2192 Name). NOT the underlying model id \u2014 deployments can be named anything you chose at create time."
         },
         {
           name: "apiVersion",
@@ -2353,7 +2353,7 @@ var azureOpenAIManifest = {
           type: "text",
           required: true,
           default: "2025-03-01-preview",
-          help: "Data-plane api-version, e.g. 2025-03-01-preview. Microsoft rolls these frequently; newer gpt-5 / o-series deployments require an api-version that supports the Responses API (2025-03-01-preview or later). See Azure OpenAI API reference docs for the current recommended version."
+          help: "Data-plane api-version. Visible in Foundry as the ?api-version=\u2026 query string in the Target URI, or the api_version=\u2026 line in the Python sample. The default 2025-03-01-preview is the minimum that works with gpt-5 / o-series deployments \u2014 Azure routes those through the Responses API internally and rejects older versions."
         }
       ]
     }
@@ -2425,7 +2425,16 @@ var AzureOpenAIAdapter = class {
       // The newer name is a superset — also accepted by gpt-4o, gpt-4-turbo,
       // and gpt-3.5-turbo deployments — so it's the safe choice for a probe
       // that has to work regardless of the underlying deployment model.
-      max_completion_tokens: 1,
+      //
+      // 256 (not 1): reasoning models consume reasoning tokens BEFORE
+      // visible output, all counted against this cap. With 1 the model
+      // can't produce any output and Azure returns
+      // `400 Could not finish the message because max_tokens or model
+      // output limit was reached`. 256 is comfortable headroom for
+      // reasoning + a "pong"-sized reply on every model we ship in the
+      // manifest, while still bounded so the probe can't run away on a
+      // misconfigured deployment.
+      max_completion_tokens: 256,
       stream: false
     });
   }
@@ -2448,6 +2457,95 @@ var azureOpenAIFactory = {
   create: () => new AzureOpenAIAdapter()
 };
 
+// src/adapters/groq/manifest.ts
+var GROQ_KNOWN_MODELS = [
+  {
+    id: "llama-3.3-70b-versatile",
+    displayName: "Llama 3.3 70B Versatile (Groq)",
+    capabilities: {
+      streaming: true,
+      tools: true,
+      vision: false,
+      systemPrompt: true,
+      maxContextTokens: 128e3
+    }
+  },
+  {
+    id: "llama-3.1-8b-instant",
+    displayName: "Llama 3.1 8B Instant (Groq)",
+    capabilities: {
+      streaming: true,
+      tools: true,
+      vision: false,
+      systemPrompt: true,
+      maxContextTokens: 128e3
+    }
+  },
+  {
+    id: "gemma2-9b-it",
+    displayName: "Gemma 2 9B (Groq)",
+    capabilities: {
+      streaming: true,
+      tools: true,
+      vision: false,
+      systemPrompt: true,
+      maxContextTokens: 8192
+    }
+  }
+];
+var groqManifest = {
+  vendor: "groq",
+  displayName: "Groq",
+  authKinds: [
+    {
+      kind: "apiKey",
+      displayName: "API Key",
+      fields: [
+        {
+          name: "apiKey",
+          label: "API Key",
+          type: "password",
+          required: true,
+          help: "Get an API key from https://console.groq.com/keys."
+        },
+        {
+          name: "baseURL",
+          label: "Base URL",
+          type: "url",
+          required: false,
+          default: "https://api.groq.com/openai/v1",
+          help: "Override only for proxies. Default points at Groq's production endpoint."
+        }
+      ]
+    }
+  ],
+  knownModels: GROQ_KNOWN_MODELS,
+  // Groq exposes /openai/v1/models with the same auth as chat completions.
+  supportsModelListing: true,
+  docsUrl: "https://console.groq.com/docs/api-reference"
+};
+
+// src/adapters/groq/adapter.ts
+var GroqAdapter = class extends OpenAIAdapter {
+  vendor = groqManifest.vendor;
+  init(auth, client) {
+    super.init(
+      {
+        ...auth,
+        baseURL: auth.baseURL ?? "https://api.groq.com/openai/v1"
+      },
+      client
+    );
+  }
+};
+
+// src/adapters/groq/factory.ts
+var groqFactory = {
+  vendor: groqManifest.vendor,
+  manifest: groqManifest,
+  create: () => new GroqAdapter()
+};
+
 // src/adapters/index.ts
 registerAdapter(openAIFactory);
 registerAdapter(deepseekFactory);
@@ -2457,6 +2555,7 @@ registerAdapter(vertexGeminiFactory);
 registerAdapter(vertexOpenAIFactory);
 registerAdapter(bedrockFactory);
 registerAdapter(azureOpenAIFactory);
+registerAdapter(groqFactory);
 
 // src/normalize-auth.ts
 function normalizeAuth(vendor, raw) {