Can AI replace Elicit?

Build the closest honest consolation tool inspired by Elicit; do not claim to replace its structural moat.
Use exactly this stack: Next.js 15 + TypeScript + SQLite + Playwright.
Primary job: Build a private AI research assistant workspace that imports user-supplied URLs or files, extracts metadata, supports notes, and searches the local corpus.
Start from an empty folder and create the complete working project.
Make the default mode single-user and private.
Store user data locally unless the core job requires the declared self-hosted database.
Do not add analytics, telemetry, ads, or third-party accounts.
Put every secret and external credential in .env and provide .env.example.
Use realistic sample data that is clearly labelled and easy to delete.
Implement the smallest polished interface that completes the core loop end to end.
Include clear empty, loading, validation, success, and failure states.
Add import and export so the user is not trapped in the app.
Use accessible keyboard navigation, labels, focus states, and sensible contrast.
Validate untrusted input and never log secrets or private file contents.
Deliberately exclude these paid-product advantages: citation graph scale; team libraries and institutional access; licensed scholarly metadata.
Do not fake integrations, network effects, proprietary data, model quality, compliance, or security claims.
Where an external API is optional, keep the app useful without it and explain the degraded mode.
Write focused unit tests for the data model and the most important workflow.
Add one end-to-end smoke test that proves the core loop works.
Create a README with setup, permissions, architecture, data location, backup, and limitations.
Add scripts for install, development, test, build, and a production-style local run.
Run the tests and build before finishing, then fix errors rather than merely describing them.