@jambonz/llm 0.5.2 → 0.5.4

package/dist/index.d.cts

@@ -31,7 +31,7 @@ interface RawCredential {
 /**
  * Convert a vendor + raw credential into an `AuthSpec` discriminated union member.
  *
- * Supported vendors: openai, deepseek, anthropic, google, vertex (alias for
+ * Supported vendors: openai, deepseek, baseten, anthropic, google, vertex (alias for
  * vertex-gemini + vertex-openai — both accept the same auth shape), vertex-gemini,
  * vertex-openai, bedrock, azure-openai, groq, huggingface.
  *

package/dist/index.d.ts

@@ -31,7 +31,7 @@ interface RawCredential {
 /**
  * Convert a vendor + raw credential into an `AuthSpec` discriminated union member.
  *
- * Supported vendors: openai, deepseek, anthropic, google, vertex (alias for
+ * Supported vendors: openai, deepseek, baseten, anthropic, google, vertex (alias for
  * vertex-gemini + vertex-openai — both accept the same auth shape), vertex-gemini,
  * vertex-openai, bedrock, azure-openai, groq, huggingface.
  *

package/dist/index.js

@@ -114,12 +114,18 @@ async function* streamFromOpenAI(client, req, options) {
   let usage;
   let finalFinishReason = null;
   let finalRawReason;
+  let sawContent = false;
   try {
     for await (const chunk of stream) {
       if (req.signal?.aborted) {
         yield { type: "end", finishReason: "aborted" };
         return;
       }
+      const chunkError = chunk.error;
+      if (chunkError) {
+        const message = chunkError && typeof chunkError === "object" && "message" in chunkError ? String(chunkError.message) : JSON.stringify(chunkError);
+        throw new Error(`OpenAI-compatible stream returned an error: ${message}`);
+      }
       if (chunk.usage) {
         usage = {
           ...chunk.usage.prompt_tokens !== void 0 ? { inputTokens: chunk.usage.prompt_tokens } : {},
@@ -133,6 +139,7 @@ async function* streamFromOpenAI(client, req, options) {
       const finishReason = choice.finish_reason;
       if (delta?.content) {
         if (t2 === void 0) t2 = process.hrtime.bigint();
+        sawContent = true;
         yield { type: "token", text: delta.content };
       }
       if (delta?.tool_calls) {
@@ -190,6 +197,12 @@ async function* streamFromOpenAI(client, req, options) {
     }
     throw err;
   }
+  const producedNothing = finalFinishReason === null && !sawContent && toolCallsEmitted.size === 0 && startedIds.size === 0 && usage === void 0;
+  if (producedNothing) {
+    throw new Error(
+      "OpenAI-compatible stream produced no data (no content, tool calls, usage, or finish reason). This usually means a misconfigured base URL or a vendor error returned in an HTTP 200 body. Verify the base URL and credentials."
+    );
+  }
   const endEvent = {
     type: "end",
     finishReason: finalFinishReason ?? "stop",
@@ -495,7 +508,7 @@ var OpenAIAdapter = class {
   }
   stream(req) {
     return streamFromOpenAI(this.ensureClient(), req, {
-      knownModels: openAIManifest.knownModels,
+      knownModels: this.knownModels(),
       vendorMetadataExtractor: OPENAI_METADATA_EXTRACTOR
     });
   }
@@ -506,7 +519,16 @@ var OpenAIAdapter = class {
     return appendOpenAIToolResult(history, toolCallId, result);
   }
   async listAvailableModels() {
-    return listOpenAICompatibleModels(this.ensureClient(), openAIManifest.knownModels);
+    return listOpenAICompatibleModels(this.ensureClient(), this.knownModels());
+  }
+  /**
+   * Capability table used to enrich both `stream()` (for tools/vision gating)
+   * and `listAvailableModels()` (for filling capability flags on live entries).
+   * Subclasses backed by the same OpenAI wire (DeepSeek, Baseten, …) override
+   * this so their own manifest entries are consulted instead of OpenAI's.
+   */
+  knownModels() {
+    return openAIManifest.knownModels;
   }
   async testCredential() {
     await this.ensureClient().models.list();
@@ -617,12 +639,311 @@ var DeepSeekAdapter = class extends OpenAIAdapter {
       client
     );
   }
+  knownModels() {
+    return deepseekManifest.knownModels;
+  }
 };
 var deepseekFactory = {
   vendor: deepseekManifest.vendor,
   manifest: deepseekManifest,
   create: () => new DeepSeekAdapter()
 };
+var basetenManifest = {
+  vendor: "baseten",
+  displayName: "Baseten",
+  authKinds: [
+    {
+      kind: "apiKey",
+      displayName: "API Key",
+      fields: [
+        {
+          name: "apiKey",
+          label: "API Key",
+          type: "password",
+          required: true,
+          help: "Get an API key from https://app.baseten.co/settings/api_keys"
+        },
+        {
+          name: "baseURL",
+          label: "Base URL",
+          type: "url",
+          required: false,
+          default: "https://inference.baseten.co/v1",
+          help: "Defaults to Baseten Model APIs. Set to https://bridge.baseten.co/v1/direct to route to a dedicated deployment."
+        }
+      ]
+    }
+  ],
+  knownModels: [
+    {
+      id: "deepseek-ai/DeepSeek-V3.1",
+      displayName: "DeepSeek V3.1",
+      capabilities: {
+        streaming: true,
+        tools: true,
+        vision: false,
+        systemPrompt: true,
+        maxContextTokens: 163840
+      }
+    },
+    {
+      id: "deepseek-ai/DeepSeek-V4-Pro",
+      displayName: "DeepSeek V4 Pro",
+      capabilities: {
+        streaming: true,
+        tools: true,
+        vision: false,
+        systemPrompt: true,
+        maxContextTokens: 131072
+      }
+    },
+    {
+      id: "zai-org/GLM-4.7",
+      displayName: "GLM 4.7",
+      capabilities: {
+        streaming: true,
+        tools: true,
+        vision: false,
+        systemPrompt: true,
+        maxContextTokens: 2e5
+      }
+    },
+    {
+      id: "zai-org/GLM-5",
+      displayName: "GLM 5",
+      capabilities: {
+        streaming: true,
+        tools: true,
+        vision: false,
+        systemPrompt: true,
+        maxContextTokens: 202800
+      }
+    },
+    {
+      id: "moonshotai/Kimi-K2.5",
+      displayName: "Kimi K2.5",
+      capabilities: {
+        streaming: true,
+        tools: true,
+        vision: true,
+        systemPrompt: true,
+        maxContextTokens: 262e3
+      }
+    },
+    {
+      id: "moonshotai/Kimi-K2.6",
+      displayName: "Kimi K2.6",
+      capabilities: {
+        streaming: true,
+        tools: true,
+        vision: true,
+        systemPrompt: true,
+        maxContextTokens: 262e3
+      }
+    },
+    {
+      id: "MiniMaxAI/MiniMax-M2.5",
+      displayName: "MiniMax M2.5",
+      capabilities: {
+        streaming: true,
+        tools: true,
+        vision: false,
+        systemPrompt: true,
+        maxContextTokens: 204e3
+      }
+    },
+    {
+      id: "nvidia/Nemotron-120B-A12B",
+      displayName: "Nemotron Super",
+      capabilities: {
+        streaming: true,
+        tools: true,
+        vision: false,
+        systemPrompt: true,
+        maxContextTokens: 202800
+      }
+    },
+    {
+      id: "openai/gpt-oss-120b",
+      displayName: "OpenAI GPT 120B",
+      capabilities: {
+        streaming: true,
+        tools: true,
+        vision: false,
+        systemPrompt: true,
+        maxContextTokens: 128072
+      }
+    }
+  ],
+  supportsModelListing: true,
+  docsUrl: "https://docs.baseten.co/inference/model-apis/overview"
+};
+var BasetenAdapter = class extends OpenAIAdapter {
+  vendor = basetenManifest.vendor;
+  init(auth, client) {
+    super.init(
+      {
+        ...auth,
+        baseURL: auth.baseURL ?? "https://inference.baseten.co/v1"
+      },
+      client
+    );
+  }
+  knownModels() {
+    return basetenManifest.knownModels;
+  }
+};
+var basetenFactory = {
+  vendor: basetenManifest.vendor,
+  manifest: basetenManifest,
+  create: () => new BasetenAdapter()
+};
+var moonshotManifest = {
+  vendor: "moonshot",
+  displayName: "Moonshot (Kimi)",
+  authKinds: [
+    {
+      kind: "apiKey",
+      displayName: "API Key",
+      fields: [
+        {
+          name: "apiKey",
+          label: "API Key",
+          type: "password",
+          required: true,
+          help: "Get an API key from https://platform.moonshot.ai/console/api-keys"
+        },
+        {
+          name: "baseURL",
+          label: "Base URL",
+          type: "url",
+          required: false,
+          default: "https://api.moonshot.ai/v1",
+          help: "Defaults to the global endpoint. Set to https://api.moonshot.cn/v1 for the China endpoint."
+        }
+      ]
+    }
+  ],
+  knownModels: [
+    {
+      id: "kimi-k2.6",
+      displayName: "Kimi K2.6",
+      capabilities: {
+        streaming: true,
+        tools: true,
+        vision: true,
+        systemPrompt: true,
+        maxContextTokens: 262144
+      }
+    },
+    {
+      id: "kimi-k2.5",
+      displayName: "Kimi K2.5",
+      capabilities: {
+        streaming: true,
+        tools: true,
+        vision: true,
+        systemPrompt: true,
+        maxContextTokens: 262144
+      }
+    }
+  ],
+  supportsModelListing: true,
+  docsUrl: "https://platform.moonshot.ai/docs"
+};
+var MoonshotAdapter = class extends OpenAIAdapter {
+  vendor = moonshotManifest.vendor;
+  init(auth, client) {
+    super.init(
+      {
+        ...auth,
+        baseURL: auth.baseURL ?? "https://api.moonshot.ai/v1"
+      },
+      client
+    );
+  }
+  knownModels() {
+    return moonshotManifest.knownModels;
+  }
+};
+var moonshotFactory = {
+  vendor: moonshotManifest.vendor,
+  manifest: moonshotManifest,
+  create: () => new MoonshotAdapter()
+};
+var zaiManifest = {
+  vendor: "zai",
+  displayName: "Z.ai (GLM)",
+  authKinds: [
+    {
+      kind: "apiKey",
+      displayName: "API Key",
+      fields: [
+        {
+          name: "apiKey",
+          label: "API Key",
+          type: "password",
+          required: true,
+          help: "Get an API key from https://z.ai/manage-apikey/apikey-list"
+        },
+        {
+          name: "baseURL",
+          label: "Base URL",
+          type: "url",
+          required: false,
+          default: "https://api.z.ai/api/paas/v4",
+          help: "Defaults to the general API. Coding-Plan keys must use https://api.z.ai/api/coding/paas/v4."
+        }
+      ]
+    }
+  ],
+  knownModels: [
+    {
+      id: "glm-5.2",
+      displayName: "GLM-5.2",
+      capabilities: {
+        streaming: true,
+        tools: true,
+        vision: false,
+        systemPrompt: true,
+        maxContextTokens: 1e6
+      }
+    },
+    {
+      id: "glm-4.7",
+      displayName: "GLM-4.7",
+      capabilities: {
+        streaming: true,
+        tools: true,
+        vision: false,
+        systemPrompt: true,
+        maxContextTokens: 2e5
+      }
+    }
+  ],
+  supportsModelListing: false,
+  docsUrl: "https://docs.z.ai/"
+};
+var ZaiAdapter = class extends OpenAIAdapter {
+  vendor = zaiManifest.vendor;
+  init(auth, client) {
+    super.init(
+      {
+        ...auth,
+        baseURL: auth.baseURL ?? "https://api.z.ai/api/paas/v4"
+      },
+      client
+    );
+  }
+  knownModels() {
+    return zaiManifest.knownModels;
+  }
+};
+var zaiFactory = {
+  vendor: zaiManifest.vendor,
+  manifest: zaiManifest,
+  create: () => new ZaiAdapter()
+};
 
 // src/adapters/anthropic/adapter.ts
 import Anthropic from "@anthropic-ai/sdk";
@@ -2801,6 +3122,9 @@ var huggingfaceFactory = {
 // src/adapters/index.ts
 registerAdapter(openAIFactory);
 registerAdapter(deepseekFactory);
+registerAdapter(basetenFactory);
+registerAdapter(moonshotFactory);
+registerAdapter(zaiFactory);
 registerAdapter(anthropicFactory);
 registerAdapter(googleFactory);
 registerAdapter(vertexGeminiFactory);
@@ -2823,6 +3147,11 @@ function normalizeAuth(vendor, raw) {
         ...requireApiKey(vendor, raw),
         baseURL: raw.api_url ?? "https://api.deepseek.com/v1"
       };
+    case "baseten":
+      return {
+        ...requireApiKey(vendor, raw),
+        baseURL: raw.api_url ?? "https://inference.baseten.co/v1"
+      };
     case "google":
       return normalizeGoogle(raw);
     case "vertex":