diff --git a/.github/workflows/phase0-qa.yml b/.github/workflows/phase0-qa.yml
index e04ecfd57ec3..0e77116e0858 100644
--- a/.github/workflows/phase0-qa.yml
+++ b/.github/workflows/phase0-qa.yml
@@ -77,6 +77,8 @@ jobs:
 
       - name: Build Electron dev app
         run: node build/lib/preLaunch.ts
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
 
       - name: Run Phase 0 QA (unit + CDP)
         run: test/cortexide-smoke/run-phase0-qa.sh --cdp
@@ -122,6 +124,8 @@ jobs:
 
       - name: Build Electron dev app
         run: node build/lib/preLaunch.ts
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
 
       - name: Run Phase 0 QA (unit + CDP)
         shell: bash
diff --git a/src/vs/workbench/contrib/cortexide/browser/react/src/settings/Settings.tsx b/src/vs/workbench/contrib/cortexide/browser/react/src/settings/Settings.tsx
index 972be0ad5055..fe226c00dc22 100644
--- a/src/vs/workbench/contrib/cortexide/browser/react/src/settings/Settings.tsx
+++ b/src/vs/workbench/contrib/cortexide/browser/react/src/settings/Settings.tsx
@@ -605,7 +605,7 @@ export const ModelDump = ({ filteredProviders }: { filteredProviders?: ProviderN
 			</div>
 		) : isAddModelOpen ? (
 			<div className="mt-4">
-				<form className="flex items-center gap-2">
+				<form className="flex flex-wrap items-center gap-2 min-w-0">
 
 					{/* Provider dropdown */}
 					<ErrorBoundary>
@@ -616,19 +616,19 @@ export const ModelDump = ({ filteredProviders }: { filteredProviders?: ProviderN
 							getOptionDisplayName={(pn) => pn ? displayInfoOfProviderName(pn).title : 'Provider Name'}
 							getOptionDropdownName={(pn) => pn ? displayInfoOfProviderName(pn).title : 'Provider Name'}
 							getOptionsEqual={(a, b) => a === b}
-							className="max-w-32 mx-2 w-full resize-none bg-void-bg-1 text-void-fg-1 placeholder:text-void-fg-3 border border-void-border-2 focus:border-void-border-1 py-1 px-2 rounded"
+							className="shrink-0 max-w-[140px] w-full resize-none bg-void-bg-1 text-void-fg-1 placeholder:text-void-fg-3 border border-void-border-2 focus:border-void-border-1 py-1 px-2 rounded"
 							arrowTouchesText={false}
 						/>
 					</ErrorBoundary>
 
-					{/* Model name input */}
+					{/* Model name input — flex so long gguf names wrap instead of horizontal scroll (onboarding #67) */}
 					<ErrorBoundary>
 						<VoidSimpleInputBox
 							value={modelName}
 							compact={true}
 							onChangeValue={setModelName}
 							placeholder='Model Name'
-							className='max-w-32'
+							className='min-w-0 flex-1 basis-40'
 						/>
 					</ErrorBoundary>
 
diff --git a/src/vs/workbench/contrib/cortexide/common/providerToolFormat.ts b/src/vs/workbench/contrib/cortexide/common/providerToolFormat.ts
index bb6af34843d7..b50c652a9744 100644
--- a/src/vs/workbench/contrib/cortexide/common/providerToolFormat.ts
+++ b/src/vs/workbench/contrib/cortexide/common/providerToolFormat.ts
@@ -124,6 +124,40 @@ export const effectiveSpecialToolFormat = (
 	isLocalInference: boolean,
 ): SpecialToolFormat => isLocalInference ? undefined : specialToolFormat
 
+/** Ollama /api/chat message — content is string; optional base64 images for vision models. */
+export type OllamaChatMessage = { role: string; content: string; images?: string[] }
+
+/** Strip data-URL prefix from base64 image payloads for Ollama's `images` field. */
+export const base64FromDataUrl = (url: string): string =>
+	url.startsWith('data:') ? url.replace(/^data:[^;]+;base64,/, '') : url
+
+/**
+ * Flatten OpenAI-format messages for Ollama's native /api/chat.
+ * Text parts join into `content`; image_url parts become base64 `images` (vision models).
+ */
+export const convertOpenAIMessagesToOllamaChat = (messages: LLMChatMessage[]): OllamaChatMessage[] =>
+	messages.map((m) => {
+		const c = (m as { content?: unknown }).content
+		let content = ''
+		const images: string[] = []
+		if (typeof c === 'string') {
+			content = c
+		} else if (Array.isArray(c)) {
+			for (const part of c) {
+				if (typeof part === 'string') { content += part }
+				else if (part?.type === 'text') { content += part.text ?? '' }
+				else if (part?.type === 'image_url' && part.image_url?.url) {
+					images.push(base64FromDataUrl(part.image_url.url))
+				}
+			}
+		} else {
+			content = c == null ? '' : String(c)
+		}
+		const out: OllamaChatMessage = { role: (m as { role: string }).role, content }
+		if (images.length) { out.images = images }
+		return out
+	})
+
 /**
  * The running accumulator for an OpenAI-compatible streaming chat response: text, reasoning, and the
  * single tool call (name / args-JSON-string / id) assembled across deltas. The OpenAI streaming
diff --git a/src/vs/workbench/contrib/cortexide/electron-main/llmMessage/sendLLMMessage.impl.ts b/src/vs/workbench/contrib/cortexide/electron-main/llmMessage/sendLLMMessage.impl.ts
index 25434ca4e7fa..1b2a8c2c7787 100644
--- a/src/vs/workbench/contrib/cortexide/electron-main/llmMessage/sendLLMMessage.impl.ts
+++ b/src/vs/workbench/contrib/cortexide/electron-main/llmMessage/sendLLMMessage.impl.ts
@@ -15,7 +15,7 @@ import { GoogleAuth } from 'google-auth-library'
 /* eslint-enable */
 
 import { GeminiLLMChatMessage, LLMChatMessage, LLMFIMMessage, ModelListParams, OllamaModelResponse, OnError, OnFinalMessage, OnText, RawToolCallObj } from '../../common/sendLLMMessageTypes.js';
-import { rawToolCallObjOfParamsStr, buildRawToolCallObj, sanitizeOpenAIMessagesForEmptyContent, toOpenAICompatibleTool, accumulateOpenAIChatDelta, buildTypedToolProperties, extractToolCallFromNonStreamingChoice, reduceGeminiChunk, finalizeGeminiToolId, effectiveSpecialToolFormat } from '../../common/providerToolFormat.js';
+import { rawToolCallObjOfParamsStr, buildRawToolCallObj, sanitizeOpenAIMessagesForEmptyContent, toOpenAICompatibleTool, accumulateOpenAIChatDelta, buildTypedToolProperties, extractToolCallFromNonStreamingChoice, reduceGeminiChunk, finalizeGeminiToolId, effectiveSpecialToolFormat, convertOpenAIMessagesToOllamaChat } from '../../common/providerToolFormat.js';
 import { formatGeminiRateLimitError } from '../../common/providerErrorFormat.js';
 import { ChatMode, displayInfoOfProviderName, FeatureName, ModelSelectionOptions, OverridesOfModel, ProviderName, SettingsOfProvider } from '../../common/cortexideSettingsTypes.js';
 import { getSendableReasoningInfo, getModelCapabilities, getProviderCapabilities, defaultProviderSettings, getReservedOutputTokenSpace } from '../../common/modelCapabilities.js';
@@ -1206,17 +1206,7 @@ const sendOllamaChat = async ({ messages, onText, onFinalMessage, onError, setti
 	}
 
 	const messagesToSend = sanitizeOpenAIMessagesForEmptyContent(messages)
-	// Ollama's native /api/chat requires messages[].content to be a STRING, but OpenAI-format messages
-	// may carry array content (multimodal text/image parts). Flatten text parts to a string (images,
-	// rare for local coding, are dropped here — text-only agentic is the target).
-	const ollamaMessages = messagesToSend.map((m) => {
-		const c = (m as any).content
-		let content: string
-		if (typeof c === 'string') { content = c }
-		else if (Array.isArray(c)) { content = c.map((part: any) => typeof part === 'string' ? part : (part?.text ?? '')).join('') }
-		else { content = c == null ? '' : String(c) }
-		return { role: (m as any).role, content }
-	})
+	const ollamaMessages = convertOpenAIMessagesToOllamaChat(messagesToSend)
 
 	let fullTextSoFar = ''
 	let firstTokenReceived = false
diff --git a/src/vs/workbench/contrib/cortexide/test/common/providerToolFormat.test.ts b/src/vs/workbench/contrib/cortexide/test/common/providerToolFormat.test.ts
index fe978c5fc5af..193e12a01923 100644
--- a/src/vs/workbench/contrib/cortexide/test/common/providerToolFormat.test.ts
+++ b/src/vs/workbench/contrib/cortexide/test/common/providerToolFormat.test.ts
@@ -5,7 +5,7 @@
 
 import * as assert from 'assert';
 import { suite, test } from 'mocha';
-import { buildRawToolCallObj, rawToolCallObjOfParamsStr, sanitizeOpenAIMessagesForEmptyContent as _sanitize, EMPTY_CONTENT_PLACEHOLDER, toOpenAICompatibleTool, accumulateOpenAIChatDelta, OpenAIChatAccumulator, OpenAIStreamDelta, effectiveSpecialToolFormat } from '../../common/providerToolFormat.js';
+import { rawToolCallObjOfParamsStr, buildRawToolCallObj, sanitizeOpenAIMessagesForEmptyContent as _sanitize, EMPTY_CONTENT_PLACEHOLDER, toOpenAICompatibleTool, accumulateOpenAIChatDelta, OpenAIChatAccumulator, OpenAIStreamDelta, effectiveSpecialToolFormat, convertOpenAIMessagesToOllamaChat, base64FromDataUrl } from '../../common/providerToolFormat.js';
 import { LLMChatMessage } from '../../common/sendLLMMessageTypes.js';
 import type { InternalToolInfo } from '../../common/prompt/prompts.js';
 
@@ -27,6 +27,30 @@ suite('effectiveSpecialToolFormat', () => {
 	});
 });
 
+suite('convertOpenAIMessagesToOllamaChat', () => {
+	test('string content passes through', () => {
+		assert.deepStrictEqual(convertOpenAIMessagesToOllamaChat([m({ role: 'user', content: 'hi' })]), [{ role: 'user', content: 'hi' }]);
+	});
+	test('text parts flatten to content string', () => {
+		assert.deepStrictEqual(
+			convertOpenAIMessagesToOllamaChat([m({ role: 'user', content: [{ type: 'text', text: 'see ' }, { type: 'text', text: 'this' }] })]),
+			[{ role: 'user', content: 'see this' }],
+		);
+	});
+	test('image_url parts become base64 images array', () => {
+		const r = convertOpenAIMessagesToOllamaChat([m({
+			role: 'user',
+			content: [{ type: 'text', text: 'what is this?' }, { type: 'image_url', image_url: { url: 'data:image/png;base64,abc123' } }],
+		})]);
+		assert.strictEqual(r[0].content, 'what is this?');
+		assert.deepStrictEqual(r[0].images, ['abc123']);
+	});
+	test('base64FromDataUrl strips data-URL prefix', () => {
+		assert.strictEqual(base64FromDataUrl('data:image/jpeg;base64,XYZ'), 'XYZ');
+		assert.strictEqual(base64FromDataUrl('rawb64'), 'rawb64');
+	});
+});
+
 suite('buildRawToolCallObj', () => {
 	test('object args -> RawToolCallObj with id/name/rawParams/doneParams/isDone', () => {
 		assert.deepStrictEqual(buildRawToolCallObj('id1', 'read_file', { uri: '/a', line: '3' }), {