Add support for Cursor transformation provider in Cursor Whisper extension

- Introduced Cursor as a new transformation provider in package.json and updated related configuration options.
- Enhanced documentation to include setup instructions for the Cursor provider and its models.
- Updated validation logic to ensure proper configuration of Cursor API keys and models.
- Added tests to cover the new Cursor provider functionality and ensure correct behavior across the extension.
- Improved user interface elements to facilitate selection and configuration of the Cursor model.

Author: efraespada

README.md

@@ -16,7 +16,7 @@ A professional VSCode/Cursor extension that captures audio from your microphone,
 1. **Install** the extension (VSIX or Marketplace when available)
 2. **Run Setup Wizard** — Command Palette → `Cursor Whisper: Setup Wizard`
 3. **Configure OpenAI API key** — Required for Whisper voice-to-text
-4. **Optionally choose optimization provider** — OpenAI, Anthropic, Google, Azure, Ollama, OpenCode, or OpenRouter
+4. **Optionally choose optimization provider** — OpenAI, Anthropic, Google, Azure, Ollama, OpenCode, OpenRouter, or Cursor
 5. **Press `Cmd+Alt+V`** and speak
 
 See the full [Quick Start Guide](docs/quickstart.md).
@@ -90,7 +90,7 @@ Developers often have complex architectural ideas, detailed requirements, or int
 ### Coming Soon
 
 - 🔄 **Prompt Transformation** - AI-powered optimization of transcribed text
-- 🔄 **Multiple AI Providers** - OpenAI, Anthropic, Google Gemini, Azure OpenAI, Ollama, OpenCode, and OpenRouter for prompt transformation
+- 🔄 **Multiple AI Providers** - OpenAI, Anthropic, Google Gemini, Azure OpenAI, Ollama, OpenCode, OpenRouter, and Cursor for prompt transformation
 - 🔄 **Chat Integration** - Direct insertion into Cursor chat input
 - 🔄 **Real-time Streaming** - See transcription as you speak
 - 🔄 **Multi-language Support** - Auto-detect or manually configure language
@@ -220,6 +220,7 @@ Prompt optimization converts transcribed speech into structured prompts. Choose
 | `ollamaBaseUrl` / `ollamaModel` | Local Ollama server settings |
 | `openCodeBaseUrl` / `openCodeModel` | Local OpenCode proxy settings |
 | `openRouterModel` | OpenRouter model (when provider is `openrouter`) |
+| `cursorModel` | Cursor model (when provider is `cursor`) |
 
 Use **Cursor Whisper: Configure Prompt Optimization Provider** to set up interactively. See [`docs/configuration/`](docs/configuration/) for provider setup.
 
@@ -229,7 +230,7 @@ Use **Cursor Whisper: Configure Prompt Optimization Provider** to set up interac
 |---------|------|---------|-------------|
 | `transcriptionLanguage` | string | `"en"` | Language for transcription (`en`, `es`, `fr`, `de`, `auto`) |
 | `enablePromptTransformation` | boolean | `true` | Transform transcription into optimized prompts |
-| `transformationProvider` | string | `"openai"` | LLM provider for transformation (`openai`, `anthropic`, `google`, `azure`, `ollama`, `opencode`, `openrouter`) |
+| `transformationProvider` | string | `"openai"` | LLM provider for transformation (`openai`, `anthropic`, `google`, `azure`, `ollama`, `opencode`, `openrouter`, `cursor`) |
 | `transformationModel` | string | `"gpt-4o"` | OpenAI model for transformation |
 | `audioQuality` | string | `"high"` | Audio recording quality (`low`, `medium`, `high`) |
 | `maxRecordingDuration` | number | `120` | Maximum recording duration in seconds |

cursor-whisper-0.1.0.vsix

@@ -36,6 +36,7 @@ Requirements:
 | Ollama | Local/offline inference |
 | OpenCode | Local multi-provider gateway via opencode-llm-proxy |
 | OpenRouter | Cloud gateway to 200+ models with one API key |
+| Cursor | Native Cursor AI models via Cursor SDK |
 
 Key aspects:
 
@@ -80,7 +81,7 @@ Key aspects:
 
 ### Negative
 
-- Increased codebase complexity (7 provider adapters)
+- Increased codebase complexity (8 provider adapters)
 - More configuration surface area for users
 - Quality and latency vary by provider
 - Additional SDK dependencies to maintain
@@ -99,6 +100,7 @@ Key aspects:
 - Value object: `src/domain/value-objects/TransformationProvider.ts`
 - Config: `cursorWhisper.transformationProvider` and provider-specific model settings
 - OpenCode and OpenRouter use the OpenAI SDK with custom `baseURL` (OpenAI-compatible APIs)
+- Cursor uses `@cursor/sdk` with `Agent.prompt()` for one-shot transformations
 - Commands: `cursor-whisper.configureTransformationProvider`, `cursor-whisper.testTransformation`
 
 ---

docs/adr/0014-multiple-transformation-providers.md

@@ -62,6 +62,7 @@ Or run **Cursor Whisper: Configure Prompt Optimization Provider**.
 | Ollama | Free | Medium | Local | Good | Privacy-first, offline |
 | OpenCode | Free | Medium | Local | High | Reuse OpenCode multi-provider setup |
 | OpenRouter | Varies | Fast | Cloud | High | 200+ models with one API key |
+| Cursor | ~$0.01 | Fast | Cloud | High | Cursor Composer and frontier models with one API key |
 
 \*Plus Whisper transcription cost (~$0.006/min, always OpenAI)
 
@@ -197,6 +198,25 @@ API keys are stored per provider (`cursor-whisper.apiKey.{provider}`). Switching
 
 ---
 
+### Option H: Cursor (SDK)
+
+```json
+{
+  "cursorWhisper.transformationProvider": "cursor",
+  "cursorWhisper.cursorModel": "composer-2.5"
+}
+```
+
+**Setup:** Configure OpenAI for Whisper first. Get a Cursor API key from [Cursor Dashboard → Integrations](https://cursor.com/dashboard/integrations). Run **Configure Prompt Optimization Provider**, select Cursor, enter your key, and choose a model.
+
+**Recommended models:** `composer-2.5` (default), `composer-2.5-fast`, `claude-4.5-sonnet`, `gpt-5.1`, `gpt-5.2-codex`
+
+**Notes:** Works in any editor (VSCode, Cursor, VSCodium, etc.). Uses the `@cursor/sdk` package to connect to Cursor's agent API. No Cursor IDE installation required — only a Cursor API key and internet access.
+
+**Pitfalls:** Cursor only handles optimization — Whisper still needs OpenAI. Ensure your Cursor account has sufficient credits.
+
+---
+
 ## Step 3: Verify Configuration
 
 Run **Cursor Whisper: Test Configuration**

docs/configuration/README.md

@@ -9,7 +9,7 @@ Get Cursor Whisper running in a few minutes.
 Cursor Whisper has **two separate services**:
 
 1. **Voice-to-text (required)** — Always uses **OpenAI Whisper**. Requires an **OpenAI API key**.
-2. **Prompt optimization (optional)** — Converts transcribed speech into structured prompts. You choose the provider (OpenAI, Anthropic, Google, Azure, Ollama, OpenCode, or OpenRouter) and supply credentials when required.
+2. **Prompt optimization (optional)** — Converts transcribed speech into structured prompts. You choose the provider (OpenAI, Anthropic, Google, Azure, Ollama, OpenCode, OpenRouter, or Cursor) and supply credentials when required.
 
 ```mermaid
 graph LR
@@ -53,7 +53,7 @@ On first launch, Cursor Whisper opens the **Setup Wizard**. You can also run it
    - Get a key: https://platform.openai.com/api-keys
 3. **Test OpenAI connection** — Verifies your key works
 4. **Enable optimization?** — Choose yes or transcription-only mode
-5. **Select provider** (if enabled) — OpenAI, Anthropic, Google, Azure, Ollama, OpenCode, or OpenRouter
+5. **Select provider** (if enabled) — OpenAI, Anthropic, Google, Azure, Ollama, OpenCode, OpenRouter, or Cursor
 6. **Provider credentials** — Enter API key or endpoint when required
 7. **Select model** — Pick the model for optimization
 8. **Test optimization** — Optional validation before finishing

docs/quickstart.md

@@ -144,7 +144,8 @@
             "azure",
             "ollama",
             "opencode",
-            "openrouter"
+            "openrouter",
+            "cursor"
           ],
           "enumDescriptions": [
             "OpenAI GPT models (requires OpenAI API key; can reuse Whisper key)",
@@ -153,7 +154,8 @@
             "Azure OpenAI deployments (requires Azure API key and endpoint)",
             "Local Ollama models (no API key required)",
             "Local OpenCode multi-provider proxy via opencode-llm-proxy (no API key required)",
-            "OpenRouter unified gateway (requires OpenRouter API key)"
+            "OpenRouter unified gateway (requires OpenRouter API key)",
+            "Native Cursor AI via Cursor SDK (works in any editor - requires Cursor API key)"
           ],
           "markdownDescription": "**Prompt optimization provider** (separate from Whisper transcription).\n\nWhisper always uses OpenAI. This setting chooses which AI service optimizes transcribed speech into structured prompts. [Configuration guide](https://github.com/vypdev/cursor-whisper/blob/master/docs/configuration/README.md)"
         },
@@ -233,6 +235,18 @@
           "default": "openai/gpt-4o",
           "markdownDescription": "**OpenRouter model identifier** for prompt optimization when `transformationProvider` is `openrouter`. Requires an OpenRouter API key."
         },
+        "cursorWhisper.cursorModel": {
+          "type": "string",
+          "default": "composer-2.5",
+          "enum": [
+            "composer-2.5",
+            "composer-2.5-fast",
+            "claude-4.5-sonnet",
+            "gpt-5.1",
+            "gpt-5.2-codex"
+          ],
+          "markdownDescription": "**Cursor model** for prompt optimization when `transformationProvider` is `cursor`. Works in any editor (VSCode, Cursor, etc.). Requires a Cursor API key from [Cursor Dashboard](https://cursor.com/dashboard/integrations)."
+        },
         "cursorWhisper.audioQuality": {
           "type": "string",
           "enum": [
@@ -283,6 +297,7 @@
   },
   "dependencies": {
     "@anthropic-ai/sdk": "^0.30.1",
+    "@cursor/sdk": "^1.0.13",
     "@google/generative-ai": "^0.21.0",
     "@kstonekuan/audio-capture": "^0.0.3",
     "@vscode/webview-ui-toolkit": "1.4.0",