# Talkform is built and running

**Open [http://localhost:3000](http://localhost:3000).**

The app includes a responsive dashboard, structured form builder, six templates, shareable voice interviews, real email/password authentication, persistent SQLite storage, transcripts, and CSV/JSON exports.

The AI interviewer collects typed answers through natural conversation, handles multiple answers at once, follows up on missing information, and accepts corrections. Respondents review their answers before submitting. Voice uses push-to-talk recording and spoken replies, with a keyboard accessibility fallback.

## API keys

An existing `GEMINI_API_KEY` was available in the environment and is connected. **No additional key is needed to try the running app.** The key stays server-side and was not copied into source files.

For another environment, set `GEMINI_API_KEY` using [.env.example](.env.example). No paid authentication, database, or separate voice service is required.

## Try it

1. Explore the private demo workspace; its sample responses are labeled.
2. Click **Try a voice interview**, or create and publish your own form.
3. Click **Create your account** to secure and keep the workspace. Sign-in and sign-out work with persistent server-side sessions.

## Verification

- **19 automated tests passed**, covering authentication, isolation, form management, validation, exports, and UI interactions.
- **Production build passed**, including TypeScript checks.
- **Live Gemini tests passed** for multi-field extraction, corrections, actual spoken-audio transcription, and submission. See [text verification](artifacts/live-ai-verification.json) and [audio verification](artifacts/live-audio-verification.json).
- Chromium was blocked by the environment’s macOS sandbox. Real-browser layout, microphone permissions, and audible playback could not be verified here; component and backend audio tests are not a substitute for those checks.

## Scope

This is a runnable MVP with a **Typeform-style structured backend**, not Typeform’s proprietary implementation or full API/feature parity. Voice is turn-based, not a continuous full-duplex call. Password recovery, email verification, billing, teams, webhooks, and native integrations are not implemented. Public production deployment still needs operational and security hardening.

See [README.md](README.md) for setup, architecture, privacy behavior, commands, and deployment considerations, and [API.md](API.md) for the backend contract. Test output is saved in [artifacts/test-results.txt](artifacts/test-results.txt); the running production server was checked in [runtime verification](artifacts/runtime-verification.json).
