ollama-ai-provider-v2 1.5.5 → 1.6.0-beta.0

package/README.md

@@ -82,11 +82,19 @@ const { text, toolCalls } = await generateText({
 Unique feature for models that support chain-of-thought reasoning:
 
 ```typescript
+// For most models - simple boolean
 const { text } = await generateText({
   model: ollama('deepseek-r1:7b'),
   providerOptions: { ollama: { think: true } },
   prompt: 'Solve this complex math problem step by step: 2x + 5 = 17',
 });
+
+// For GPT-OSS models - use thinking levels for trace length control
+const { text } = await generateText({
+  model: ollama('gpt-oss:7b'),
+  providerOptions: { ollama: { think: 'high' } }, // 'low', 'medium', or 'high'
+  prompt: 'Analyze this complex scenario in detail',
+});
 ```
 
 ### Advanced Ollama Options
@@ -174,7 +182,7 @@ Works with any model in your Ollama installation, including:
 
 - **Chat Models**: `llama3.2`, `mistral`, `phi4-mini`, `qwen2.5`, `codellama`, `gemma3`
 - **Vision Models**: `llava`, `llama3.2-vision`, `minicpm-v`
-- **Reasoning Models**: `deepseek-r1:7b`, `deepseek-r1:1.5b`, `deepseek-r1:8b`
+- **Reasoning Models**: `deepseek-r1:7b`, `deepseek-r1:1.5b`, `deepseek-r1:8b`, `gpt-oss:7b`
 - **Code Models**: `codellama:code`, `codellama:python`, `deepseek-coder-v2`
 - **Embedding Models**: `nomic-embed-text`, `all-minilm`, `mxbai-embed-large`
 

package/dist/index.d.mts

@@ -4,7 +4,11 @@ import { z } from 'zod/v4';
 
 type OllamaChatModelId = "athene-v2" | "athene-v2:72b" | "aya-expanse" | "aya-expanse:8b" | "aya-expanse:32b" | "codegemma" | "codegemma:2b" | "codegemma:7b" | "codellama" | "codellama:7b" | "codellama:13b" | "codellama:34b" | "codellama:70b" | "codellama:code" | "codellama:python" | "command-r" | "command-r:35b" | "command-r-plus" | "command-r-plus:104b" | "command-r7b" | "command-r7b:7b" | "deepseek-r1" | "deepseek-r1:1.5b" | "deepseek-r1:7b" | "deepseek-r1:8b" | "deepseek-r1:14b" | "deepseek-r1:32b" | "deepseek-r1:70b" | "deepseek-r1:671b" | "deepseek-coder-v2" | "deepseek-coder-v2:16b" | "deepseek-coder-v2:236b" | "deepseek-v3" | "deepseek-v3:671b" | "devstral" | "devstral:24b" | "dolphin3" | "dolphin3:8b" | "exaone3.5" | "exaone3.5:2.4b" | "exaone3.5:7.8b" | "exaone3.5:32b" | "falcon2" | "falcon2:11b" | "falcon3" | "falcon3:1b" | "falcon3:3b" | "falcon3:7b" | "falcon3:10b" | "firefunction-v2" | "firefunction-v2:70b" | "gemma" | "gemma:2b" | "gemma:7b" | "gemma2" | "gemma2:2b" | "gemma2:9b" | "gemma2:27b" | "gemma3" | "gemma3:1b" | "gemma3:4b" | "gemma3:12b" | "gemma3:27b" | "granite3-dense" | "granite3-dense:2b" | "granite3-dense:8b" | "granite3-guardian" | "granite3-guardian:2b" | "granite3-guardian:8b" | "granite3-moe" | "granite3-moe:1b" | "granite3-moe:3b" | "granite3.1-dense" | "granite3.1-dense:2b" | "granite3.1-dense:8b" | "granite3.1-moe" | "granite3.1-moe:1b" | "granite3.1-moe:3b" | "llama2" | "llama2:7b" | "llama2:13b" | "llama2:70b" | "llama3" | "llama3:8b" | "llama3:70b" | "llama3-chatqa" | "llama3-chatqa:8b" | "llama3-chatqa:70b" | "llama3-gradient" | "llama3-gradient:8b" | "llama3-gradient:70b" | "llama3.1" | "llama3.1:8b" | "llama3.1:70b" | "llama3.1:405b" | "llama3.2" | "llama3.2:1b" | "llama3.2:3b" | "llama3.2-vision" | "llama3.2-vision:11b" | "llama3.2-vision:90b" | "llama3.3" | "llama3.3:70b" | "llama4" | "llama4:16x17b" | "llama4:128x17b" | "llama-guard3" | "llama-guard3:1b" | "llama-guard3:8b" | "llava" | "llava:7b" | "llava:13b" | "llava:34b" | "llava-llama3" | "llava-llama3:8b" | "llava-phi3" | "llava-phi3:3.8b" | "marco-o1" | "marco-o1:7b" | "mistral" | "mistral:7b" | "mistral-large" | "mistral-large:123b" | "mistral-nemo" | "mistral-nemo:12b" | "mistral-small" | "mistral-small:22b" | "mixtral" | "mixtral:8x7b" | "mixtral:8x22b" | "moondream" | "moondream:1.8b" | "openhermes" | "openhermes:v2.5" | "nemotron" | "nemotron:70b" | "nemotron-mini" | "nemotron-mini:4b" | "olmo" | "olmo:7b" | "olmo:13b" | "opencoder" | "opencoder:1.5b" | "opencoder:8b" | "phi3" | "phi3:3.8b" | "phi3:14b" | "phi3.5" | "phi3.5:3.8b" | "phi4" | "phi4:14b" | "qwen" | "qwen:7b" | "qwen:14b" | "qwen:32b" | "qwen:72b" | "qwen:110b" | "qwen2" | "qwen2:0.5b" | "qwen2:1.5b" | "qwen2:7b" | "qwen2:72b" | "qwen2.5" | "qwen2.5:0.5b" | "qwen2.5:1.5b" | "qwen2.5:3b" | "qwen2.5:7b" | "qwen2.5:14b" | "qwen2.5:32b" | "qwen2.5:72b" | "qwen2.5-coder" | "qwen2.5-coder:0.5b" | "qwen2.5-coder:1.5b" | "qwen2.5-coder:3b" | "qwen2.5-coder:7b" | "qwen2.5-coder:14b" | "qwen2.5-coder:32b" | "qwen3" | "qwen3:0.6b" | "qwen3:1.7b" | "qwen3:4b" | "qwen3:8b" | "qwen3:14b" | "qwen3:30b" | "qwen3:32b" | "qwen3:235b" | "qwq" | "qwq:32b" | "sailor2" | "sailor2:1b" | "sailor2:8b" | "sailor2:20b" | "shieldgemma" | "shieldgemma:2b" | "shieldgemma:9b" | "shieldgemma:27b" | "smallthinker" | "smallthinker:3b" | "smollm" | "smollm:135m" | "smollm:360m" | "smollm:1.7b" | "tinyllama" | "tinyllama:1.1b" | "tulu3" | "tulu3:8b" | "tulu3:70b" | (string & {});
 declare const ollamaProviderOptions: z.ZodObject<{
-    think: z.ZodOptional<z.ZodBoolean>;
+    think: z.ZodOptional<z.ZodUnion<readonly [z.ZodBoolean, z.ZodEnum<{
+        low: "low";
+        medium: "medium";
+        high: "high";
+    }>]>>;
     options: z.ZodOptional<z.ZodObject<{
         num_ctx: z.ZodOptional<z.ZodNumber>;
         repeat_last_n: z.ZodOptional<z.ZodNumber>;
@@ -27,9 +31,12 @@ interface OllamaCompletionSettings {
      * the model's thinking from the model's output. When disabled, the model will not think
      * and directly output the content.
      *
-     * Only supported by certain models like DeepSeek R1 and Qwen 3.
+     * For GPT-OSS models: accepts "low", "medium", or "high" to tune trace length.
+     * For other models: accepts boolean true/false.
+     *
+     * Only supported by certain models like DeepSeek R1, Qwen 3, and GPT-OSS.
      */
-    think?: boolean;
+    think?: boolean | "low" | "medium" | "high";
     /**
      * Echo back the prompt in addition to the completion.
      */
@@ -146,7 +153,11 @@ declare const ollamaEmbeddingProviderOptions: z.ZodObject<{
 type OllamaEmbeddingProviderOptions = z.infer<typeof ollamaEmbeddingProviderOptions>;
 
 declare const ollamaCompletionProviderOptions: z.ZodObject<{
-    think: z.ZodOptional<z.ZodBoolean>;
+    think: z.ZodOptional<z.ZodUnion<readonly [z.ZodBoolean, z.ZodEnum<{
+        low: "low";
+        medium: "medium";
+        high: "high";
+    }>]>>;
     user: z.ZodOptional<z.ZodString>;
     suffix: z.ZodOptional<z.ZodString>;
     echo: z.ZodOptional<z.ZodBoolean>;

package/dist/index.d.ts

@@ -4,7 +4,11 @@ import { z } from 'zod/v4';
 
 type OllamaChatModelId = "athene-v2" | "athene-v2:72b" | "aya-expanse" | "aya-expanse:8b" | "aya-expanse:32b" | "codegemma" | "codegemma:2b" | "codegemma:7b" | "codellama" | "codellama:7b" | "codellama:13b" | "codellama:34b" | "codellama:70b" | "codellama:code" | "codellama:python" | "command-r" | "command-r:35b" | "command-r-plus" | "command-r-plus:104b" | "command-r7b" | "command-r7b:7b" | "deepseek-r1" | "deepseek-r1:1.5b" | "deepseek-r1:7b" | "deepseek-r1:8b" | "deepseek-r1:14b" | "deepseek-r1:32b" | "deepseek-r1:70b" | "deepseek-r1:671b" | "deepseek-coder-v2" | "deepseek-coder-v2:16b" | "deepseek-coder-v2:236b" | "deepseek-v3" | "deepseek-v3:671b" | "devstral" | "devstral:24b" | "dolphin3" | "dolphin3:8b" | "exaone3.5" | "exaone3.5:2.4b" | "exaone3.5:7.8b" | "exaone3.5:32b" | "falcon2" | "falcon2:11b" | "falcon3" | "falcon3:1b" | "falcon3:3b" | "falcon3:7b" | "falcon3:10b" | "firefunction-v2" | "firefunction-v2:70b" | "gemma" | "gemma:2b" | "gemma:7b" | "gemma2" | "gemma2:2b" | "gemma2:9b" | "gemma2:27b" | "gemma3" | "gemma3:1b" | "gemma3:4b" | "gemma3:12b" | "gemma3:27b" | "granite3-dense" | "granite3-dense:2b" | "granite3-dense:8b" | "granite3-guardian" | "granite3-guardian:2b" | "granite3-guardian:8b" | "granite3-moe" | "granite3-moe:1b" | "granite3-moe:3b" | "granite3.1-dense" | "granite3.1-dense:2b" | "granite3.1-dense:8b" | "granite3.1-moe" | "granite3.1-moe:1b" | "granite3.1-moe:3b" | "llama2" | "llama2:7b" | "llama2:13b" | "llama2:70b" | "llama3" | "llama3:8b" | "llama3:70b" | "llama3-chatqa" | "llama3-chatqa:8b" | "llama3-chatqa:70b" | "llama3-gradient" | "llama3-gradient:8b" | "llama3-gradient:70b" | "llama3.1" | "llama3.1:8b" | "llama3.1:70b" | "llama3.1:405b" | "llama3.2" | "llama3.2:1b" | "llama3.2:3b" | "llama3.2-vision" | "llama3.2-vision:11b" | "llama3.2-vision:90b" | "llama3.3" | "llama3.3:70b" | "llama4" | "llama4:16x17b" | "llama4:128x17b" | "llama-guard3" | "llama-guard3:1b" | "llama-guard3:8b" | "llava" | "llava:7b" | "llava:13b" | "llava:34b" | "llava-llama3" | "llava-llama3:8b" | "llava-phi3" | "llava-phi3:3.8b" | "marco-o1" | "marco-o1:7b" | "mistral" | "mistral:7b" | "mistral-large" | "mistral-large:123b" | "mistral-nemo" | "mistral-nemo:12b" | "mistral-small" | "mistral-small:22b" | "mixtral" | "mixtral:8x7b" | "mixtral:8x22b" | "moondream" | "moondream:1.8b" | "openhermes" | "openhermes:v2.5" | "nemotron" | "nemotron:70b" | "nemotron-mini" | "nemotron-mini:4b" | "olmo" | "olmo:7b" | "olmo:13b" | "opencoder" | "opencoder:1.5b" | "opencoder:8b" | "phi3" | "phi3:3.8b" | "phi3:14b" | "phi3.5" | "phi3.5:3.8b" | "phi4" | "phi4:14b" | "qwen" | "qwen:7b" | "qwen:14b" | "qwen:32b" | "qwen:72b" | "qwen:110b" | "qwen2" | "qwen2:0.5b" | "qwen2:1.5b" | "qwen2:7b" | "qwen2:72b" | "qwen2.5" | "qwen2.5:0.5b" | "qwen2.5:1.5b" | "qwen2.5:3b" | "qwen2.5:7b" | "qwen2.5:14b" | "qwen2.5:32b" | "qwen2.5:72b" | "qwen2.5-coder" | "qwen2.5-coder:0.5b" | "qwen2.5-coder:1.5b" | "qwen2.5-coder:3b" | "qwen2.5-coder:7b" | "qwen2.5-coder:14b" | "qwen2.5-coder:32b" | "qwen3" | "qwen3:0.6b" | "qwen3:1.7b" | "qwen3:4b" | "qwen3:8b" | "qwen3:14b" | "qwen3:30b" | "qwen3:32b" | "qwen3:235b" | "qwq" | "qwq:32b" | "sailor2" | "sailor2:1b" | "sailor2:8b" | "sailor2:20b" | "shieldgemma" | "shieldgemma:2b" | "shieldgemma:9b" | "shieldgemma:27b" | "smallthinker" | "smallthinker:3b" | "smollm" | "smollm:135m" | "smollm:360m" | "smollm:1.7b" | "tinyllama" | "tinyllama:1.1b" | "tulu3" | "tulu3:8b" | "tulu3:70b" | (string & {});
 declare const ollamaProviderOptions: z.ZodObject<{
-    think: z.ZodOptional<z.ZodBoolean>;
+    think: z.ZodOptional<z.ZodUnion<readonly [z.ZodBoolean, z.ZodEnum<{
+        low: "low";
+        medium: "medium";
+        high: "high";
+    }>]>>;
     options: z.ZodOptional<z.ZodObject<{
         num_ctx: z.ZodOptional<z.ZodNumber>;
         repeat_last_n: z.ZodOptional<z.ZodNumber>;
@@ -27,9 +31,12 @@ interface OllamaCompletionSettings {
      * the model's thinking from the model's output. When disabled, the model will not think
      * and directly output the content.
      *
-     * Only supported by certain models like DeepSeek R1 and Qwen 3.
+     * For GPT-OSS models: accepts "low", "medium", or "high" to tune trace length.
+     * For other models: accepts boolean true/false.
+     *
+     * Only supported by certain models like DeepSeek R1, Qwen 3, and GPT-OSS.
      */
-    think?: boolean;
+    think?: boolean | "low" | "medium" | "high";
     /**
      * Echo back the prompt in addition to the completion.
      */
@@ -146,7 +153,11 @@ declare const ollamaEmbeddingProviderOptions: z.ZodObject<{
 type OllamaEmbeddingProviderOptions = z.infer<typeof ollamaEmbeddingProviderOptions>;
 
 declare const ollamaCompletionProviderOptions: z.ZodObject<{
-    think: z.ZodOptional<z.ZodBoolean>;
+    think: z.ZodOptional<z.ZodUnion<readonly [z.ZodBoolean, z.ZodEnum<{
+        low: "low";
+        medium: "medium";
+        high: "high";
+    }>]>>;
     user: z.ZodOptional<z.ZodString>;
     suffix: z.ZodOptional<z.ZodString>;
     echo: z.ZodOptional<z.ZodBoolean>;

package/dist/index.js

@@ -31,7 +31,7 @@ var import_provider_utils8 = require("@ai-sdk/provider-utils");
 
 // src/completion/ollama-completion-language-model.ts
 var import_provider_utils2 = require("@ai-sdk/provider-utils");
-var import_v42 = require("zod/v4");
+var import_v43 = require("zod/v4");
 
 // src/completion/ollama-error.ts
 var import_v4 = require("zod/v4");
@@ -156,12 +156,63 @@ function getResponseMetadata({
   };
 }
 
+// src/ollama-chat-settings.ts
+var import_v42 = require("zod/v4");
+var ollamaProviderOptions = import_v42.z.object({
+  /**
+   * Enable or disable the model's thinking process. When enabled, the output will separate
+   * the model's thinking from the model's output. When disabled, the model will not think
+   * and directly output the content.
+   *
+   * For GPT-OSS models: accepts "low", "medium", or "high" to tune trace length.
+   * For other models: accepts boolean true/false.
+   * 
+   * Only supported by certain models like DeepSeek R1, Qwen 3, and GPT-OSS.
+   */
+  think: import_v42.z.union([
+    import_v42.z.boolean(),
+    import_v42.z.enum(["low", "medium", "high"])
+  ]).optional(),
+  options: import_v42.z.object({
+    num_ctx: import_v42.z.number().optional(),
+    repeat_last_n: import_v42.z.number().optional(),
+    repeat_penalty: import_v42.z.number().optional(),
+    temperature: import_v42.z.number().optional(),
+    seed: import_v42.z.number().optional(),
+    stop: import_v42.z.array(import_v42.z.string()).optional(),
+    num_predict: import_v42.z.number().optional(),
+    top_k: import_v42.z.number().optional(),
+    top_p: import_v42.z.number().optional(),
+    min_p: import_v42.z.number().optional()
+  }).optional()
+});
+function validateThinkParameter(modelId, think) {
+  if (think === void 0) {
+    return void 0;
+  }
+  const isGptOss = modelId.toLowerCase().includes("gpt-oss");
+  if (isGptOss) {
+    if (typeof think === "boolean") {
+      return think ? "medium" : void 0;
+    }
+    return think;
+  } else {
+    if (typeof think === "string") {
+      return think !== void 0;
+    }
+    return think;
+  }
+}
+
 // src/completion/ollama-completion-language-model.ts
-var ollamaCompletionProviderOptions = import_v42.z.object({
-  think: import_v42.z.boolean().optional(),
-  user: import_v42.z.string().optional(),
-  suffix: import_v42.z.string().optional(),
-  echo: import_v42.z.boolean().optional()
+var ollamaCompletionProviderOptions = import_v43.z.object({
+  think: import_v43.z.union([
+    import_v43.z.boolean(),
+    import_v43.z.enum(["low", "medium", "high"])
+  ]).optional(),
+  user: import_v43.z.string().optional(),
+  suffix: import_v43.z.string().optional(),
+  echo: import_v43.z.boolean().optional()
 });
 var OllamaCompletionLanguageModel = class {
   constructor(modelId, settings, config) {
@@ -219,13 +270,14 @@ var OllamaCompletionLanguageModel = class {
     }
     const { prompt: completionPrompt, stopSequences } = convertToOllamaCompletionPrompt({ prompt });
     const stop = [...stopSequences != null ? stopSequences : [], ...userStopSequences != null ? userStopSequences : []];
+    const validatedThink = validateThinkParameter(this.modelId, ollamaOptions.think);
     return {
       args: {
         // model id:
         model: this.modelId,
         // Ollama-supported settings:
         user: ollamaOptions.user,
-        think: ollamaOptions.think,
+        think: validatedThink,
         // standardized settings:
         max_tokens: maxOutputTokens,
         temperature,
@@ -365,28 +417,28 @@ var OllamaCompletionLanguageModel = class {
     };
   }
 };
-var baseOllamaResponseSchema = import_v42.z.object({
-  model: import_v42.z.string(),
-  created_at: import_v42.z.string(),
-  response: import_v42.z.string(),
-  done: import_v42.z.boolean(),
-  context: import_v42.z.array(import_v42.z.number()),
-  eval_count: import_v42.z.number().optional(),
-  eval_duration: import_v42.z.number().optional(),
-  load_duration: import_v42.z.number().optional(),
-  total_duration: import_v42.z.number().optional(),
-  prompt_eval_count: import_v42.z.number().optional(),
-  prompt_eval_duration: import_v42.z.number().optional()
+var baseOllamaResponseSchema = import_v43.z.object({
+  model: import_v43.z.string(),
+  created_at: import_v43.z.string(),
+  response: import_v43.z.string(),
+  done: import_v43.z.boolean(),
+  context: import_v43.z.array(import_v43.z.number()),
+  eval_count: import_v43.z.number().optional(),
+  eval_duration: import_v43.z.number().optional(),
+  load_duration: import_v43.z.number().optional(),
+  total_duration: import_v43.z.number().optional(),
+  prompt_eval_count: import_v43.z.number().optional(),
+  prompt_eval_duration: import_v43.z.number().optional()
 });
 
 // src/embedding/ollama-embedding-model.ts
 var import_provider2 = require("@ai-sdk/provider");
 var import_provider_utils3 = require("@ai-sdk/provider-utils");
-var import_v43 = require("zod/v4");
-var ollamaEmbeddingProviderOptions = import_v43.z.object({
-  dimensions: import_v43.z.number().optional(),
-  truncate: import_v43.z.boolean().optional(),
-  keepAlive: import_v43.z.string().optional()
+var import_v44 = require("zod/v4");
+var ollamaEmbeddingProviderOptions = import_v44.z.object({
+  dimensions: import_v44.z.number().optional(),
+  truncate: import_v44.z.boolean().optional(),
+  keepAlive: import_v44.z.string().optional()
 });
 var OllamaEmbeddingModel = class {
   constructor(modelId, settings, config) {
@@ -468,12 +520,12 @@ var OllamaEmbeddingModel = class {
     };
   }
 };
-var ollamaTextEmbeddingResponseSchema = import_v43.z.object({
-  model: import_v43.z.string(),
-  embeddings: import_v43.z.array(import_v43.z.array(import_v43.z.number())),
-  total_duration: import_v43.z.number(),
-  load_duration: import_v43.z.number(),
-  prompt_eval_count: import_v43.z.number()
+var ollamaTextEmbeddingResponseSchema = import_v44.z.object({
+  model: import_v44.z.string(),
+  embeddings: import_v44.z.array(import_v44.z.array(import_v44.z.number())),
+  total_duration: import_v44.z.number(),
+  load_duration: import_v44.z.number(),
+  prompt_eval_count: import_v44.z.number()
 });
 
 // src/responses/ollama-responses-language-model.ts
@@ -805,31 +857,6 @@ function prepareResponsesTools({
   }
 }
 
-// src/ollama-chat-settings.ts
-var import_v44 = require("zod/v4");
-var ollamaProviderOptions = import_v44.z.object({
-  /**
-   * Enable or disable the model's thinking process. When enabled, the output will separate
-   * the model's thinking from the model's output. When disabled, the model will not think
-   * and directly output the content.
-   *
-   * Only supported by certain models like DeepSeek R1 and Qwen 3.
-   */
-  think: import_v44.z.boolean().optional(),
-  options: import_v44.z.object({
-    num_ctx: import_v44.z.number().optional(),
-    repeat_last_n: import_v44.z.number().optional(),
-    repeat_penalty: import_v44.z.number().optional(),
-    temperature: import_v44.z.number().optional(),
-    seed: import_v44.z.number().optional(),
-    stop: import_v44.z.array(import_v44.z.string()).optional(),
-    num_predict: import_v44.z.number().optional(),
-    top_k: import_v44.z.number().optional(),
-    top_p: import_v44.z.number().optional(),
-    min_p: import_v44.z.number().optional()
-  }).optional()
-});
-
 // src/responses/ollama-responses-request-builder.ts
 var OllamaRequestBuilder = class {
   async buildRequest({
@@ -922,7 +949,8 @@ var OllamaRequestBuilder = class {
     responseFormat,
     ollamaOptions
   }) {
-    var _a, _b;
+    var _a;
+    const validatedThink = validateThinkParameter(modelId, ollamaOptions == null ? void 0 : ollamaOptions.think);
     return {
       model: modelId,
       messages: convertToOllamaChatMessages({
@@ -935,8 +963,8 @@ var OllamaRequestBuilder = class {
       ...(responseFormat == null ? void 0 : responseFormat.type) === "json" && {
         format: responseFormat.schema != null ? responseFormat.schema : "json"
       },
-      think: (_a = ollamaOptions == null ? void 0 : ollamaOptions.think) != null ? _a : false,
-      options: (_b = ollamaOptions == null ? void 0 : ollamaOptions.options) != null ? _b : void 0
+      think: validatedThink,
+      options: (_a = ollamaOptions == null ? void 0 : ollamaOptions.options) != null ? _a : void 0
     };
   }
 };