Can I vibecode Tower?

Build a highly usable, production-ready personal replacement for the core loop of the Tower Git client in an empty repository.
Use exactly this tech stack without deviation: Python 3.12, Typer for the CLI framework, and SQLite for local state management. Do not suggest or offer alternative languages or frameworks.

Your primary objective is to build a repository-local Git client assistant that acts as a powerful local agent. It must index a single target codebase, interface with one specified LLM provider, propose intelligent diffs based on user prompts, automatically run test suites to verify changes, and meticulously record every modification applied to the repository.

Core Features & Workflows:
- Start from an empty folder and scaffold the complete working CLI project.
- Make the default mode strictly single-user and private to the local machine.
- Store all user configuration, indexing metadata, and interaction history locally using the declared SQLite database.
- Keep all secrets (e.g., API keys for the LLM) in a `.env` file, provide a `.env.example`, and ensure no secrets or private file contents are ever logged or transmitted beyond the single model API call.
- Use realistic, clearly labelled sample data during setup that is easy for the user to delete or ignore.
- Implement the smallest but most polished CLI interface possible that guides the user smoothly through the core loop: prompt -> index -> propose -> test -> commit.
- Include crystal-clear empty states (e.g., when no history exists), loading spinners during API calls, rigorous input validation, and distinct success/failure outputs.
- Add an export command so the user's interaction history and saved preferences can be easily extracted in JSON format.
- Ensure accessible terminal navigation, clear color-coding for diffs, and sensible contrast in the output.

Explicit Exclusions & Constraints:
- Do NOT add analytics, telemetry, advertisements, or third-party user accounts.
- Deliberately exclude features that rely on large-context cloud infrastructure or proprietary vector databases.
- Exclude IDE-wide integrations, focusing solely on the repository-local CLI experience.
- Exclude enterprise policy enforcement, telemetry dashboards, and official support channels.
- Do NOT fake or simulate integrations, network effects, proprietary indexing algorithms, or compliance certifications.
- If the configured external LLM API is unavailable, the application must degrade gracefully, remaining useful as an advanced local Git wrapper and explaining the degraded mode clearly to the user.

Testing & Documentation:
- Write focused unit tests covering the SQLite data model, the file indexing logic, and the prompt-generation workflow.
- Add at least one comprehensive end-to-end smoke test that proves the core loop (proposing a diff and recording it) works correctly against a dummy Git repository.
- Create an exhaustive README containing setup instructions, permission requirements, architectural overview, data location details, backup steps, and known limitations.
- Provide helper scripts (e.g., Make or bash) for installation, development, testing, building, and running a production-style local instance.

Finalize your work by running the tests and build steps before finishing. If errors occur, fix them in code rather than merely describing what went wrong.