ollama-ai-provider-v2 1.6.0-beta.0 → 2.0.0-beta.0

package/README.md

@@ -7,6 +7,20 @@ Use Ollama with the Vercel AI SDK, implementing the official Ollama API. This pr
 [![Node.js](https://img.shields.io/badge/Node.js-18+-green.svg)](https://nodejs.org/)
 [![License: Apache-2.0](https://img.shields.io/badge/License-Apache--2.0-yellow.svg)](https://opensource.org/licenses/Apache-2.0)
 
+## 🎉 AI SDK 6 Beta Support
+
+This provider now supports **AI SDK 6 Beta** features including:
+
+- **🤖 Agent Abstraction** - Build complex agents with `ToolLoopAgent`
+- **🔐 Tool Approval** - Request user confirmation before executing tools  
+- **📊 Structured Output** - Generate typed data alongside tool calling
+- **⚡ Enhanced Performance** - Optimized for the latest AI SDK features
+
+```bash
+# Install AI SDK 6 Beta + Ollama Provider
+npm install ai@beta ollama-ai-provider-v2
+```
+
 ## Why Choose Ollama Provider V2?
 
 - ✅ **Minimal Dependencies** - Lean codebase with just 2 core dependencies
@@ -82,19 +96,11 @@ 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
@@ -182,7 +188,7 @@ 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`, `gpt-oss:7b`
+- **Reasoning Models**: `deepseek-r1:7b`, `deepseek-r1:1.5b`, `deepseek-r1:8b`
 - **Code Models**: `codellama:code`, `codellama:python`, `deepseek-coder-v2`
 - **Embedding Models**: `nomic-embed-text`, `all-minilm`, `mxbai-embed-large`
 
@@ -200,6 +206,147 @@ ollama serve
 ollama pull llama3.2
 ```
 
+# AI SDK 6 Beta examples
+## Agent Abstraction
+
+AI SDK 6 introduces the `ToolLoopAgent` class for building agents with full control over execution flow.
+
+### Basic Agent
+
+```typescript
+import { ToolLoopAgent } from 'ai';
+import { ollama } from 'ollama-ai-provider-v2';
+
+const weatherAgent = new ToolLoopAgent({
+  model: ollama('llama3.3:70b'),
+  instructions: 'You are a helpful weather assistant.',
+  tools: {
+    weather: weatherTool,
+  },
+});
+
+const result = await weatherAgent.generate({
+  prompt: 'What is the weather in San Francisco?',
+});
+```
+
+### Agent with Call Options
+
+Use call options to pass runtime configuration to agents:
+
+```typescript
+import { ToolLoopAgent } from 'ai';
+import { ollama } from 'ollama-ai-provider-v2';
+import { z } from 'zod';
+
+const supportAgent = new ToolLoopAgent({
+  model: ollama('qwen2.5:32b'),
+  callOptionsSchema: z.object({
+    userId: z.string(),
+    accountType: z.enum(['free', 'pro', 'enterprise']),
+  }),
+  instructions: 'You are a helpful customer support agent.',
+  prepareCall: ({ options, ...settings }) => ({
+    ...settings,
+    instructions: `${settings.instructions}
+
+User context:
+- Account type: ${options.accountType}  
+- User ID: ${options.userId}
+
+Adjust your response based on the user's account level.`,
+  }),
+});
+
+const result = await supportAgent.generate({
+  prompt: 'How do I upgrade my account?',
+  options: {
+    userId: 'user_123',
+    accountType: 'free',
+  },
+});
+```
+
+## Tool Execution Approval
+
+AI SDK 6 allows you to require user approval before executing tools.
+
+### Basic Tool Approval
+
+```typescript
+import { tool } from 'ai';
+import { z } from 'zod';
+
+export const weatherTool = tool({
+  description: 'Get the weather in a location',
+  inputSchema: z.object({
+    city: z.string(),
+  }),
+  needsApproval: true, // Always require approval
+  execute: async ({ city }) => {
+    const weather = await fetchWeather(city);
+    return weather;
+  },
+});
+```
+
+### Dynamic Approval
+
+Make approval decisions based on tool input:
+
+```typescript
+export const paymentTool = tool({
+  description: 'Process a payment',
+  inputSchema: z.object({
+    amount: z.number(),
+    recipient: z.string(),
+  }),
+  needsApproval: async ({ amount }) => amount > 1000, // Only large payments
+  execute: async ({ amount, recipient }) => {
+    return await processPayment(amount, recipient);
+  },
+});
+```
+
+## UI Integration
+
+### Server-side API Route
+
+```typescript
+import { createAgentUIStreamResponse } from 'ai';
+import { weatherAgent } from '@/lib/agents';
+
+export async function POST(request: Request) {
+  const { messages } = await request.json();
+
+  return createAgentUIStreamResponse({
+    agent: weatherAgent,
+    messages,
+  });
+}
+```
+
+### Client-side with Type Safety
+
+```typescript
+import { useChat } from '@ai-sdk/react';
+import { InferAgentUIMessage } from 'ai';
+import { weatherAgent } from '@/lib/agents';
+
+type WeatherAgentUIMessage = InferAgentUIMessage<typeof weatherAgent>;
+
+export function WeatherChat() {
+  const { messages, sendMessage } = useChat<WeatherAgentUIMessage>();
+
+  return (
+    <div>
+      {/* Your chat UI */}
+    </div>
+  );
+}
+```
+
+
 ## Contributing
 
 Contributions are welcome! Here's how to get started:

package/dist/index.d.mts

@@ -4,11 +4,7 @@ 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.ZodUnion<readonly [z.ZodBoolean, z.ZodEnum<{
-        low: "low";
-        medium: "medium";
-        high: "high";
-    }>]>>;
+    think: z.ZodOptional<z.ZodBoolean>;
     options: z.ZodOptional<z.ZodObject<{
         num_ctx: z.ZodOptional<z.ZodNumber>;
         repeat_last_n: z.ZodOptional<z.ZodNumber>;
@@ -31,12 +27,9 @@ interface OllamaCompletionSettings {
      * 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.
+     * Only supported by certain models like DeepSeek R1 and Qwen 3.
      */
-    think?: boolean | "low" | "medium" | "high";
+    think?: boolean;
     /**
      * Echo back the prompt in addition to the completion.
      */
@@ -153,11 +146,7 @@ declare const ollamaEmbeddingProviderOptions: z.ZodObject<{
 type OllamaEmbeddingProviderOptions = z.infer<typeof ollamaEmbeddingProviderOptions>;
 
 declare const ollamaCompletionProviderOptions: z.ZodObject<{
-    think: z.ZodOptional<z.ZodUnion<readonly [z.ZodBoolean, z.ZodEnum<{
-        low: "low";
-        medium: "medium";
-        high: "high";
-    }>]>>;
+    think: z.ZodOptional<z.ZodBoolean>;
     user: z.ZodOptional<z.ZodString>;
     suffix: z.ZodOptional<z.ZodString>;
     echo: z.ZodOptional<z.ZodBoolean>;

package/dist/index.d.ts

@@ -4,11 +4,7 @@ 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.ZodUnion<readonly [z.ZodBoolean, z.ZodEnum<{
-        low: "low";
-        medium: "medium";
-        high: "high";
-    }>]>>;
+    think: z.ZodOptional<z.ZodBoolean>;
     options: z.ZodOptional<z.ZodObject<{
         num_ctx: z.ZodOptional<z.ZodNumber>;
         repeat_last_n: z.ZodOptional<z.ZodNumber>;
@@ -31,12 +27,9 @@ interface OllamaCompletionSettings {
      * 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.
+     * Only supported by certain models like DeepSeek R1 and Qwen 3.
      */
-    think?: boolean | "low" | "medium" | "high";
+    think?: boolean;
     /**
      * Echo back the prompt in addition to the completion.
      */
@@ -153,11 +146,7 @@ declare const ollamaEmbeddingProviderOptions: z.ZodObject<{
 type OllamaEmbeddingProviderOptions = z.infer<typeof ollamaEmbeddingProviderOptions>;
 
 declare const ollamaCompletionProviderOptions: z.ZodObject<{
-    think: z.ZodOptional<z.ZodUnion<readonly [z.ZodBoolean, z.ZodEnum<{
-        low: "low";
-        medium: "medium";
-        high: "high";
-    }>]>>;
+    think: z.ZodOptional<z.ZodBoolean>;
     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_v43 = require("zod/v4");
+var import_v42 = require("zod/v4");
 
 // src/completion/ollama-error.ts
 var import_v4 = require("zod/v4");
@@ -156,63 +156,38 @@ 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;
+// src/completion/ollama-completion-language-model.ts
+function createJsonStreamResponseHandler(schema) {
+  return async ({ response }) => {
+    if (!response.body) {
+      throw new Error("Response body is null");
     }
-    return think;
-  }
+    const stream = response.body.pipeThrough(new TextDecoderStream()).pipeThrough(new TransformStream({
+      transform(chunk, controller) {
+        const lines = chunk.split("\n");
+        for (const line of lines) {
+          if (line.trim()) {
+            const result = (0, import_provider_utils2.safeParseJSON)({ text: line.trim(), schema });
+            controller.enqueue(result);
+          }
+        }
+      }
+    }));
+    const responseHeaders = {};
+    response.headers.forEach((value, key) => {
+      responseHeaders[key] = value;
+    });
+    return {
+      value: stream,
+      responseHeaders
+    };
+  };
 }
-
-// src/completion/ollama-completion-language-model.ts
-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 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 OllamaCompletionLanguageModel = class {
   constructor(modelId, settings, config) {
@@ -270,14 +245,13 @@ 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: validatedThink,
+        think: ollamaOptions.think,
         // standardized settings:
         max_tokens: maxOutputTokens,
         temperature,
@@ -356,7 +330,7 @@ var OllamaCompletionLanguageModel = class {
       headers: (0, import_provider_utils2.combineHeaders)(this.config.headers(), options.headers),
       body,
       failedResponseHandler: ollamaFailedResponseHandler,
-      successfulResponseHandler: (0, import_provider_utils2.createJsonStreamResponseHandler)(
+      successfulResponseHandler: createJsonStreamResponseHandler(
         baseOllamaResponseSchema
       ),
       abortSignal: options.abortSignal,
@@ -417,28 +391,28 @@ var OllamaCompletionLanguageModel = class {
     };
   }
 };
-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()
+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()
 });
 
 // src/embedding/ollama-embedding-model.ts
 var import_provider2 = require("@ai-sdk/provider");
 var import_provider_utils3 = require("@ai-sdk/provider-utils");
-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 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 OllamaEmbeddingModel = class {
   constructor(modelId, settings, config) {
@@ -520,12 +494,12 @@ var OllamaEmbeddingModel = class {
     };
   }
 };
-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()
+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()
 });
 
 // src/responses/ollama-responses-language-model.ts
@@ -857,6 +831,31 @@ 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({
@@ -949,8 +948,7 @@ var OllamaRequestBuilder = class {
     responseFormat,
     ollamaOptions
   }) {
-    var _a;
-    const validatedThink = validateThinkParameter(modelId, ollamaOptions == null ? void 0 : ollamaOptions.think);
+    var _a, _b;
     return {
       model: modelId,
       messages: convertToOllamaChatMessages({
@@ -963,8 +961,8 @@ var OllamaRequestBuilder = class {
       ...(responseFormat == null ? void 0 : responseFormat.type) === "json" && {
         format: responseFormat.schema != null ? responseFormat.schema : "json"
       },
-      think: validatedThink,
-      options: (_a = ollamaOptions == null ? void 0 : ollamaOptions.options) != null ? _a : void 0
+      think: (_a = ollamaOptions == null ? void 0 : ollamaOptions.think) != null ? _a : false,
+      options: (_b = ollamaOptions == null ? void 0 : ollamaOptions.options) != null ? _b : void 0
     };
   }
 };
@@ -1264,6 +1262,32 @@ var OllamaStreamProcessor = class {
 };
 
 // src/responses/ollama-responses-language-model.ts
+function createJsonStreamResponseHandler2(schema) {
+  return async ({ response }) => {
+    if (!response.body) {
+      throw new Error("Response body is null");
+    }
+    const stream = response.body.pipeThrough(new TextDecoderStream()).pipeThrough(new TransformStream({
+      transform(chunk, controller) {
+        const lines = chunk.split("\n");
+        for (const line of lines) {
+          if (line.trim()) {
+            const result = (0, import_provider_utils7.safeParseJSON)({ text: line.trim(), schema });
+            controller.enqueue(result);
+          }
+        }
+      }
+    }));
+    const responseHeaders = {};
+    response.headers.forEach((value, key) => {
+      responseHeaders[key] = value;
+    });
+    return {
+      value: stream,
+      responseHeaders
+    };
+  };
+}
 var OllamaResponsesLanguageModel = class {
   constructor(modelId, config) {
     this.specificationVersion = "v2";
@@ -1321,7 +1345,7 @@ var OllamaResponsesLanguageModel = class {
       headers: (0, import_provider_utils7.combineHeaders)(this.config.headers(), options.headers),
       body: { ...body, stream: true },
       failedResponseHandler: ollamaFailedResponseHandler,
-      successfulResponseHandler: (0, import_provider_utils7.createJsonStreamResponseHandler)(baseOllamaResponseSchema2),
+      successfulResponseHandler: createJsonStreamResponseHandler2(baseOllamaResponseSchema2),
       abortSignal: options.abortSignal,
       fetch: this.config.fetch
     });