@jambonz/llm 0.6.5 → 0.6.7

package/dist/index.d.cts

@@ -34,7 +34,7 @@ interface RawCredential {
  * Supported vendors: openai, deepseek, baseten, moonshot, zai, xai, bedrock-mantle,
  * anthropic, google, vertex (alias for vertex-gemini + vertex-openai — both accept
  * the same auth shape), vertex-gemini, vertex-openai, bedrock, azure-openai, groq,
- * huggingface.
+ * huggingface, openrouter.
  *
  * Throws with a clear message if the raw credential is missing required fields.
  */

package/dist/index.d.ts

@@ -34,7 +34,7 @@ interface RawCredential {
  * Supported vendors: openai, deepseek, baseten, moonshot, zai, xai, bedrock-mantle,
  * anthropic, google, vertex (alias for vertex-gemini + vertex-openai — both accept
  * the same auth shape), vertex-gemini, vertex-openai, bedrock, azure-openai, groq,
- * huggingface.
+ * huggingface, openrouter.
  *
  * Throws with a clear message if the raw credential is missing required fields.
  */

package/dist/index.js

@@ -142,12 +142,20 @@ async function* streamFromOpenAI(client, req, options) {
         throw new Error(`OpenAI-compatible stream returned an error: ${message}`);
       }
       if (chunk.usage) {
-        const cachedTokens = chunk.usage.prompt_tokens_details?.cached_tokens;
+        const details = chunk.usage.prompt_tokens_details;
+        const cachedTokens = details?.cached_tokens;
+        const cacheWriteTokens = details?.cache_write_tokens;
         usage = {
-          ...chunk.usage.prompt_tokens !== void 0 ? { inputTokens: chunk.usage.prompt_tokens - (cachedTokens ?? 0) } : {},
+          ...chunk.usage.prompt_tokens !== void 0 ? {
+            inputTokens: Math.max(
+              0,
+              chunk.usage.prompt_tokens - (cachedTokens ?? 0) - (cacheWriteTokens ?? 0)
+            )
+          } : {},
           ...chunk.usage.completion_tokens !== void 0 ? { outputTokens: chunk.usage.completion_tokens } : {},
           ...chunk.usage.total_tokens !== void 0 ? { totalTokens: chunk.usage.total_tokens } : {},
-          ...cachedTokens !== void 0 ? { cacheReadTokens: cachedTokens } : {}
+          ...cachedTokens !== void 0 ? { cacheReadTokens: cachedTokens } : {},
+          ...cacheWriteTokens !== void 0 ? { cacheWriteTokens } : {}
         };
       }
       const choice = chunk.choices?.[0];
@@ -3544,6 +3552,179 @@ var huggingfaceFactory = {
   create: () => new HuggingfaceAdapter()
 };
 
+// src/adapters/openrouter/manifest.ts
+var OPENROUTER_KNOWN_MODELS = [
+  {
+    id: "anthropic/claude-haiku-5.5",
+    displayName: "Claude Haiku 5.5 (OpenRouter)",
+    capabilities: {
+      streaming: true,
+      tools: true,
+      vision: true,
+      systemPrompt: true,
+      maxContextTokens: 1e6
+    }
+  },
+  {
+    id: "anthropic/claude-sonnet-5.5",
+    displayName: "Claude Sonnet 5.5 (OpenRouter)",
+    capabilities: {
+      streaming: true,
+      tools: true,
+      vision: true,
+      systemPrompt: true,
+      maxContextTokens: 1e6
+    }
+  },
+  {
+    id: "openai/gpt-5.4-mini",
+    displayName: "GPT-5.4 Mini (OpenRouter)",
+    capabilities: {
+      streaming: true,
+      tools: true,
+      vision: true,
+      systemPrompt: true,
+      maxContextTokens: 4e5
+    }
+  },
+  {
+    id: "openai/gpt-4.1-mini",
+    displayName: "GPT-4.1 Mini (OpenRouter)",
+    capabilities: {
+      streaming: true,
+      tools: true,
+      vision: true,
+      systemPrompt: true,
+      maxContextTokens: 1047576
+    }
+  },
+  {
+    id: "google/gemini-3.5-flash",
+    displayName: "Gemini 3.5 Flash (OpenRouter)",
+    capabilities: {
+      streaming: true,
+      tools: true,
+      vision: true,
+      systemPrompt: true,
+      maxContextTokens: 1048576
+    }
+  },
+  {
+    id: "meta-llama/llama-3.3-70b-instruct",
+    displayName: "Llama 3.3 70B Instruct (OpenRouter)",
+    capabilities: {
+      streaming: true,
+      tools: true,
+      vision: false,
+      systemPrompt: true,
+      maxContextTokens: 131072
+    }
+  }
+];
+var openRouterManifest = {
+  vendor: "openrouter",
+  displayName: "OpenRouter",
+  authKinds: [
+    {
+      kind: "apiKey",
+      displayName: "API Key",
+      fields: [
+        {
+          name: "apiKey",
+          label: "API Key",
+          type: "password",
+          required: true,
+          help: "Get an API key from https://openrouter.ai/settings/keys"
+        },
+        {
+          name: "baseURL",
+          label: "Base URL",
+          type: "url",
+          required: false,
+          default: "https://openrouter.ai/api/v1",
+          help: "Override only for proxies. Default points at OpenRouter's production endpoint."
+        }
+      ]
+    }
+  ],
+  knownModels: OPENROUTER_KNOWN_MODELS,
+  // /models is public; the adapter enriches it with OpenRouter's own capability metadata.
+  supportsModelListing: true,
+  docsUrl: "https://openrouter.ai/docs/api_reference/overview"
+};
+
+// src/adapters/openrouter/adapter.ts
+var OPENROUTER_METADATA_EXTRACTOR = makeMetadataExtractor([
+  { header: "x-generation-id", key: "request_id" }
+]);
+var OPENROUTER_ATTRIBUTION_HEADERS = {
+  "HTTP-Referer": "https://jambonz.org",
+  "X-OpenRouter-Title": "jambonz"
+};
+var OpenRouterAdapter = class extends OpenAIAdapter {
+  vendor = openRouterManifest.vendor;
+  init(auth, client) {
+    super.init(
+      {
+        ...auth,
+        baseURL: auth.baseURL ?? "https://openrouter.ai/api/v1"
+      },
+      client
+    );
+    this.client = this.ensureClient().withOptions({
+      defaultHeaders: OPENROUTER_ATTRIBUTION_HEADERS
+    });
+  }
+  stream(req) {
+    return streamFromOpenAI(this.ensureClient(), req, {
+      knownModels: this.knownModels(),
+      vendorMetadataExtractor: OPENROUTER_METADATA_EXTRACTOR,
+      // Verified accepted by OpenRouter (forwarded to OpenAI-family upstreams).
+      includeCacheKey: true,
+      openRouterAutoCache: this.isOpenRouter
+    });
+  }
+  async listAvailableModels() {
+    const result = await this.ensureClient().models.list();
+    const out = [];
+    for (const m of result.data) {
+      if (m.id.endsWith(":batch")) continue;
+      const outputs = m.architecture?.output_modalities;
+      if (outputs && !outputs.includes("text")) continue;
+      const known = this.knownModels().find((k) => k.id === m.id);
+      if (known) {
+        out.push(known);
+        continue;
+      }
+      out.push({
+        id: m.id,
+        ...m.name ? { displayName: m.name } : {},
+        capabilities: {
+          streaming: true,
+          tools: m.supported_parameters?.includes("tools") ?? true,
+          vision: m.architecture?.input_modalities?.includes("image") ?? false,
+          systemPrompt: true,
+          ...m.context_length ? { maxContextTokens: m.context_length } : {}
+        }
+      });
+    }
+    return out;
+  }
+  async testCredential() {
+    await this.ensureClient().get("/key");
+  }
+  knownModels() {
+    return openRouterManifest.knownModels;
+  }
+};
+
+// src/adapters/openrouter/factory.ts
+var openRouterFactory = {
+  vendor: openRouterManifest.vendor,
+  manifest: openRouterManifest,
+  create: () => new OpenRouterAdapter()
+};
+
 // src/adapters/index.ts
 registerAdapter(openAIFactory);
 registerAdapter(deepseekFactory);
@@ -3560,6 +3741,7 @@ registerAdapter(bedrockFactory);
 registerAdapter(azureOpenAIFactory);
 registerAdapter(groqFactory);
 registerAdapter(huggingfaceFactory);
+registerAdapter(openRouterFactory);
 
 // src/normalize-auth.ts
 function normalizeAuth(vendor, raw) {
@@ -3568,6 +3750,7 @@ function normalizeAuth(vendor, raw) {
     case "anthropic":
     case "groq":
     case "huggingface":
+    case "openrouter":
     case "moonshot":
     case "zai":
     case "xai":