v0.0.47

- API key is used (if needed and provided) on getting the list of models for adding OpenAI compatible provider
- Multiline field for Edit with AI
- Qwen3.5 models (2B, 4B, 9B) added in the predefined list  - good for tools and chat

package.json

@@ -2,7 +2,7 @@
   "name": "llama-vscode",
   "displayName": "llama-vscode",
   "description": "Local LLM-assisted text completion using llama.cpp",
-  "version": "0.0.45",
+  "version": "0.0.47",
   "publisher": "ggml-org",
   "repository": "https://github.com/ggml-org/llama.vscode",
   "engines": {

src/lists.ts

@@ -79,7 +79,36 @@ export const PREDEFINED_LISTS = new Map<string, any>([
               "endpoint": "http://127.0.0.1:8010"
             }
           ]],
-[PREDEFINED_LISTS_KEYS.TOOLS,  [
+[PREDEFINED_LISTS_KEYS.TOOLS,
+    [
+            {
+              "name": "Qwen3.5-2B-GGUF:Q8_0 (LOCAL) (CPU)",
+              "localStartCommand": "llama-server -hf unsloth/Qwen3.5-2B-GGUF:Q8_0 --jinja  -c 0 -ub 1024 -b 1024 --cache-reuse 256 --port 8009 --host 127.0.0.1",
+              "endpoint": "http://localhost:8009",
+              "aiModel": "",
+              "isKeyRequired": false
+            },
+            {
+              "name": "Qwen3.5-2B-GGUF:Q8_0 (LOCAL) (VRAM>3GB)",
+              "localStartCommand": "llama-server -hf unsloth/Qwen3.5-2B-GGUF:Q8_0 --jinja -ngl 99  -c 0 -ub 1024 -b 1024 --cache-reuse 256 --port 8009 --host 127.0.0.1",
+              "endpoint": "http://localhost:8009",
+              "aiModel": "",
+              "isKeyRequired": false
+            },
+            {
+              "name": "Qwen3.5-4B-GGUF:Q8_0 (LOCAL) (VRAM>6GB)",
+              "localStartCommand": "llama-server -hf unsloth/Qwen3.5-4B-GGUF:Q8_0 --jinja -c 0 -ub 1024 -b 1024 --cache-reuse 256 --port 8009 --host 127.0.0.1",
+              "endpoint": "http://localhost:8009",
+              "aiModel": "",
+              "isKeyRequired": false
+            },
+            {
+              "name": "Qwen3.5-9B-GGUF:Q8_0 (LOCAL) (VRAM>12GB)",
+              "localStartCommand": "llama-server -hf unsloth/Qwen3.5-9B-GGUF:Q8_0 --jinja -c 0 -ub 1024 -b 1024 --cache-reuse 256 --port 8009 --host 127.0.0.1",
+              "endpoint": "http://localhost:8009",
+              "aiModel": "",
+              "isKeyRequired": false
+            },
             {
               "name": "OpenAI gpt-oss 20B (LOCAL) (> 19GB VRAM)",
               "localStartCommand": "llama-server -hf ggml-org/gpt-oss-20b-GGUF -c 0 --jinja --reasoning-format none -np 2 --port 8009",

src/services/openai-comp-model-strategy.ts

@@ -44,13 +44,13 @@ export class OpenAiCompModelStrategy implements IAddStrategy {
                     prompt: 'example: http://localhost:8080 or https://openrauter.ai/api'
                 })??""
                 isKeyRequired = await Utils.confirmAction(`Is API key required for this endpoint (${endpoint})?`, "");
-            } 
+            }
             if (!endpoint){
                 vscode.window.showWarningMessage("Endpoint is not provided!")
                 return;
             }
             const providerModels: QuickPickItem[] = [];
-            const models = await this.getModels(endpoint);
+            const models = await this.getModels(endpoint, isKeyRequired);
             if (models.length == 0) {
                 vscode.window.showInformationMessage("No models are found.")
                 return
@@ -108,30 +108,50 @@ export class OpenAiCompModelStrategy implements IAddStrategy {
         }
     }
 
-    private async getModels(endpoint: string): Promise<OpenAiCompModel[]> {
-        let hfEndpoint = Utils.trimTrailingSlash(endpoint) +"/v1/models";
+    private async getModels(endpoint: string, isKeyRequired: boolean): Promise<OpenAiCompModel[]> {
+        const hfEndpoint = Utils.trimTrailingSlash(endpoint) + "/v1/models";
+
+        // Create a request configuration
+        let requestConfig: any = {};
+
+        if (isKeyRequired) {
+            // We get the saved key for this specific endpoint
+            const apiKey = this.app.persistence.getApiKey(endpoint);
+            if (apiKey) {
+                requestConfig = {
+                    headers: {
+                        'Authorization': `Bearer ${apiKey}`,
+                        'Content-Type': 'application/json'
+                    }
+                };
+            }
+        }
+
         try {
-            let result = await axios.default.get(
-                `${Utils.trimTrailingSlash(hfEndpoint)}`
+            const result = await axios.default.get(
+                `${Utils.trimTrailingSlash(hfEndpoint)}`,
+                requestConfig
             );
+
             let models: OpenAiCompModel[] = [];
-        
-            let modelsList: OpenAiCompModel[] = []
-            if (result && result.data && result.data.models) modelsList = result.data.models
-            else if (result && result.data && result.data.data) modelsList = result.data.data
-            if (modelsList.length > 0){
-                for(let mdl of modelsList){
-                    models.push(mdl)
+            let modelsList: OpenAiCompModel[] = [];
+
+            if (result && result.data && result.data.models) modelsList = result.data.models;
+            else if (result && result.data && result.data.data) modelsList = result.data.data;
+
+            if (modelsList.length > 0) {
+                for (let mdl of modelsList) {
+                    models.push(mdl);
                 }
             }
-            
+
             return models;
-        } catch (error){
-            vscode.window.showErrorMessage("Error getting provider models): " + error)
+        } catch (error) {
+            vscode.window.showErrorMessage("Error getting provider models: " + error);
             return [];
         }
     }
-    
+
     private sanitizeInput(input: string): string {
         return input ? input.trim() : '';
     }

src/text-editor.ts

@@ -26,6 +26,192 @@ export class TextEditor {
         vscode.commands.executeCommand('setContext', 'textEditSuggestionVisible', visible);
     }
 
+    private escapeWebviewAttr(value: string): string {
+        return value
+            .replace(/&/g, '&')
+            .replace(/"/g, '"')
+            .replace(/'/g, ''')
+            .replace(/</g, '&lt;');
+    }
+
+    /**
+     * Multiline instructions (webview); resolves undefined if cancelled or closed.
+     */
+    private showMultilineEditPrompt(): Promise<string | undefined> {
+        const title =
+            this.app.configuration.getUiText('How would you like to modify the selected text?') ??
+            'How would you like to modify the selected text?';
+        const placeholder =
+            this.app.configuration.getUiText('Enter your instructions for editing the text...') ??
+            'Enter your instructions for editing the text...';
+        const submitLabel = this.app.configuration.getUiText('Submit') ?? 'Submit';
+        const cancelLabel = this.app.configuration.getUiText('Cancel') ?? 'Cancel';
+        const emptyHint =
+            this.app.configuration.getUiText('Please enter editing instructions.') ??
+            'Please enter editing instructions.';
+
+        return new Promise((resolve) => {
+            let settled = false;
+            const panel = vscode.window.createWebviewPanel(
+                'editWithAiMultilinePrompt',
+                title,
+                { viewColumn: vscode.ViewColumn.Beside, preserveFocus: false },
+                { enableScripts: true }
+            );
+
+            const finish = (value: string | undefined) => {
+                if (settled) {
+                    return;
+                }
+                settled = true;
+                resolve(value);
+                panel.dispose();
+            };
+
+            const cspSource = panel.webview.cspSource;
+            panel.webview.html = `<!DOCTYPE html>
+<html lang="en">
+<head>
+    <meta charset="UTF-8">
+    <meta http-equiv="Content-Security-Policy" content="default-src 'none'; style-src ${cspSource} 'unsafe-inline'; script-src 'unsafe-inline' ${cspSource};">
+    <meta name="viewport" content="width=device-width, initial-scale=1.0">
+    <style>
+        body {
+            box-sizing: border-box;
+            margin: 0;
+            padding: 12px;
+            height: 100vh;
+            display: flex;
+            flex-direction: column;
+            font-family: var(--vscode-font-family);
+            font-size: var(--vscode-font-size);
+            color: var(--vscode-foreground);
+            background-color: var(--vscode-editor-background);
+        }
+        label {
+            margin-bottom: 8px;
+        }
+        textarea {
+            flex: 1;
+            min-height: 120px;
+            resize: vertical;
+            padding: 8px;
+            border: 1px solid var(--vscode-input-border);
+            background: var(--vscode-input-background);
+            color: var(--vscode-input-foreground);
+            font-family: var(--vscode-editor-font-family);
+            font-size: var(--vscode-editor-font-size);
+        }
+        textarea:focus {
+            outline: 1px solid var(--vscode-focusBorder);
+        }
+        .actions {
+            margin-top: 12px;
+            display: flex;
+            gap: 8px;
+            justify-content: flex-end;
+        }
+        /* DOM order is Submit then Cancel (Tab: textarea → Submit → Cancel); flex order keeps Cancel left, Submit right. */
+        .actions .secondary {
+            order: 1;
+        }
+        .actions .primary {
+            order: 2;
+        }
+        button {
+            padding: 6px 14px;
+            border: none;
+            cursor: pointer;
+            font-size: var(--vscode-font-size);
+        }
+        .primary {
+            background: var(--vscode-button-background);
+            color: var(--vscode-button-foreground);
+        }
+        .primary:hover {
+            background: var(--vscode-button-hoverBackground);
+        }
+        .secondary {
+            background: var(--vscode-button-secondaryBackground);
+            color: var(--vscode-button-secondaryForeground);
+        }
+        .secondary:hover {
+            background: var(--vscode-button-secondaryHoverBackground);
+        }
+    </style>
+</head>
+<body>
+    <label for="prompt">${this.escapeWebviewAttr(title)}</label>
+    <textarea id="prompt" placeholder="${this.escapeWebviewAttr(placeholder)}" autofocus></textarea>
+    <div class="actions">
+        <button type="button" class="primary" id="submit">${this.escapeWebviewAttr(submitLabel)}</button>
+        <button type="button" class="secondary" id="cancel">${this.escapeWebviewAttr(cancelLabel)}</button>
+    </div>
+    <script>
+        const vscode = acquireVsCodeApi();
+        const ta = document.getElementById('prompt');
+        function focusPrompt() {
+            if (!ta) {
+                return;
+            }
+            ta.focus();
+            const len = ta.value.length;
+            ta.setSelectionRange(len, len);
+        }
+        window.addEventListener('load', focusPrompt);
+        requestAnimationFrame(focusPrompt);
+        setTimeout(focusPrompt, 0);
+        setTimeout(focusPrompt, 100);
+        window.addEventListener('message', (event) => {
+            const data = event.data;
+            if (data && data.command === 'focusPrompt') {
+                focusPrompt();
+            }
+        });
+        document.getElementById('submit').addEventListener('click', () => {
+            vscode.postMessage({ command: 'submit', text: ta.value });
+        });
+        document.getElementById('cancel').addEventListener('click', () => {
+            vscode.postMessage({ command: 'cancel' });
+        });
+    </script>
+</body>
+</html>`;
+
+            const requestPromptFocus = () => {
+                void panel.webview.postMessage({ command: 'focusPrompt' });
+            };
+            panel.onDidChangeViewState((e) => {
+                if (e.webviewPanel.visible) {
+                    requestPromptFocus();
+                }
+            });
+            requestPromptFocus();
+            setTimeout(requestPromptFocus, 50);
+            setTimeout(requestPromptFocus, 200);
+
+            panel.webview.onDidReceiveMessage((message) => {
+                if (message.command === 'submit') {
+                    const text = typeof message.text === 'string' ? message.text : '';
+                    if (!text.trim()) {
+                        void vscode.window.showInformationMessage(emptyHint);
+                        return;
+                    }
+                    finish(text);
+                } else if (message.command === 'cancel') {
+                    finish(undefined);
+                }
+            });
+
+            panel.onDidDispose(() => {
+                if (!settled) {
+                    settled = true;
+                    resolve(undefined);
+                }
+            });
+        });
+    }
+
     async showEditPrompt(editor: vscode.TextEditor) {
         let chatUrl = this.app.configuration.endpoint_chat
         if (!chatUrl) chatUrl = this.app.configuration.endpoint_tools; 
@@ -64,12 +250,7 @@ export class TextEditor {
         const contextRange = new vscode.Range(startLine, 0, endLine, editor.document.lineAt(endLine).text.length);
         const context = editor.document.getText(contextRange);
         
-        // Create and show input box
-        const prompt = await vscode.window.showInputBox({
-            placeHolder: 'Enter your instructions for editing the text...',
-            prompt: 'How would you like to modify the selected text?',
-            ignoreFocusOut: true
-        });
+        const prompt = await this.showMultilineEditPrompt();
 
         if (!prompt) {
             return;