Can I vibecode HyperWrite?

Build a detailed, self-hosted clone inspired by HyperWrite focusing on its core AI writing capabilities. Use exactly this stack: Next.js 15 (App Router), TypeScript, Tailwind CSS, SQLite, Drizzle ORM, and the OpenAI API.

**DATA MODEL & SCHEMA SPECIFICS**
Implement the following core entities in SQLite with Drizzle ORM:
- `documents`: `id`, `title`, `content` (text), `created_at`, `updated_at`.
- `templates`: `id`, `name`, `system_prompt` (text), `tone` (string).
- `snippets`: `id`, `document_id`, `text`, `source_url`, `created_at`.
- `revisions`: `id`, `document_id`, `previous_content`, `new_content`, `created_at`.
Ensure all relations have cascading deletes where appropriate. Use foreign key constraints.

**SCREENS & UX PATTERNS**
1. **Dashboard (`/dashboard`)**: A grid layout of recent documents and a prominent "New Document from Template" button.
2. **Editor (`/editor/[id]`)**: A rich-text editor (using ProseMirror or similar) taking up the main screen. Features inline "ghost text" for AI autocomplete suggestions that the user can accept with the 'Tab' key.
3. **AI Sidebar**: A collapsible right sidebar containing a chat interface for brainstorming, generating outlines, and querying the document context.
4. **Rewrite Toolkit**: A floating toolbar that appears when text is highlighted, offering actions like "Shorten", "Expand", "Formalize", and "Fix Grammar".

**STEP-BY-STEP USER JOURNEYS**
- **Journey 1 (Briefing)**: User creates a new document, fills out a brief form (audience, objective, tone, source URLs), and the app generates an initial outline.
- **Journey 2 (Drafting)**: User approves the outline, and the AI drafts the first section. The user starts typing, and the AI continuously provides inline completions.
- **Journey 3 (Refining)**: User highlights a paragraph, clicks "Rewrite" in the floating toolbar, and selects "Make it punchier". The AI streams the replacement text inline.
- **Journey 4 (Export)**: User finishes editing and clicks "Export", downloading a clean Markdown file with a companion JSON research bundle.

**TECHNICAL SPECIFICATIONS & PROTOCOLS**
- Use Server Actions for database mutations.
- Integrate the OpenAI API using streaming responses (Server-Sent Events) for real-time text generation in the editor and sidebar.
- Implement debounce logic for keystrokes to optimize AI autocomplete API calls.
- Use SQLite FTS5 for chunking and retrieving imported source text.
- Data export functionality: Provide a utility to generate standard Markdown files.

Start from an empty folder, create the complete project, and ensure the default mode is single-user and private. Do not add analytics or telemetry. Put all secrets in `.env`. Validate untrusted input with Zod. Write unit tests for the core logic and one end-to-end smoke test using Playwright. Run the tests and build before finishing.