Can I vibecode Circleback?

text
Build a deliberately narrow personal substitute for Circleback, focusing strictly on a private, local-first meeting notes pipeline, and do not attempt to create a full clone. You must use exactly this technology stack: Python 3.12, FastAPI for the backend API, whisper.cpp for local audio transcription, and SQLite for structured data storage. Do not introduce any alternative frameworks or databases.

Your primary objective is to create a complete working project from an empty folder that accepts an uploaded audio file, transcribes the speech to text using whisper.cpp, processes the transcription into structured meeting notes (e.g., summary, action items, key decisions), and provides a Markdown export functionality.

The application must be single-user and private by default. All user data, including uploaded audio files, transcriptions, and generated notes, must be stored locally using the SQLite database and local file system. Do not include any analytics, telemetry, advertising, or third-party account integrations. Any required external credentials (e.g., for an optional LLM API used to summarize the transcript, if applicable) must be stored in a .env file, and you must provide a .env.example template. Never hardcode secrets.

Provide realistic, clearly labeled sample data that is easy for the user to delete or ignore. Implement a polished but minimalistic frontend interface (using vanilla HTML/JS or a simple templating engine if needed) that successfully completes the core loop from file upload to note generation and export. This interface must include clear visual states for empty data, loading/processing, input validation, success, and failure/error conditions. The user must not be trapped; ensure both import and export capabilities are robust.

Ensure the UI is fully accessible, featuring keyboard navigation, clear aria-labels, distinct focus states, and sensible color contrast. Strongly validate all untrusted input and ensure that secrets or private file contents are never logged to the console or log files.

You must deliberately exclude paid-product advantages found in Circleback, such as cross-call team analytics, automated meeting-bot auto-join capabilities, and live multi-speaker real-time accuracy. Do not fake integrations, network effects, proprietary data, model quality, compliance, or security claims. If any external API is optional, the app must remain fully functional in a degraded mode without it, and this must be explained in the UI.

Write highly focused unit tests for the SQLite data model and the main transcription/summarization workflow. Include at least one end-to-end smoke test that proves the core loop works from start to finish. Provide a comprehensive README containing setup instructions, permissions required, architecture overview, data location, backup procedures, and known limitations. Add scripts for install, development, test, build, and production-style local run. Run the tests and build before finishing, and fix any errors rather than merely describing them.