Write it.
Hear it back.

Lector is a writing environment with read-aloud built in. Write your draft, then hear it sentence by sentence. The voice catches what the eye skips over. Completely local. No cloud. No subscription.

Every writer knows the feeling. You read a paragraph three times and it looks fine. Then you hear it read aloud and immediately catch the sentence that runs too long, the word that appears twice in four lines, the transition that doesn’t quite land. That is not a grammar problem. It is a prose problem. The eye misses it because the brain already knows what you meant to say.

Hearing your writing is different. The voice does not know what you intended. It reads what is actually there.

Lector puts read-aloud inside the writing environment itself. No copying text into another application. No browser tabs. No switching tools mid-edit. You write, you listen, you revise, all in the same place.

The editor is distraction-free. WYSIWYG mode renders formatting as you type. Raw Markdown is available for those who want it. You toggle between them without losing your place. Nothing else on screen.

Read-aloud has twelve built-in offline voices via Kokoro, and it pairs every sentence with word-by-word highlighting so the playhead is always visible. Click any word to jump there. The mechanic is mundane and the effect is not: most writers who use read-aloud tools never go back to editing in silence.

Upload five to fifteen seconds of audio and Lector builds a voice profile from it using F5-TTS. After that, every read-aloud session can use your own voice, or a voice familiar to you. The profile is built once and lives on your computer permanently.

Voice cloning runs on your GPU. There is no upload, no remote API, no third-party service. The reference clip is processed locally, the model runs locally, and the resulting profile is stored locally. If you uninstall Lector, the profile goes with it.

Local-first is half the privacy story. The other half is what sits on disk after you close the editor. Lector’s library writes documents to disk as age v1 ciphertext under a key only your operating system knows about. Plaintext export stays one click away when you want it.

The encryption is age v1 with X25519 keys and ChaCha20-Poly1305 ciphers. The plaintext document never sits in the library directory, even briefly: a scratch directory mediates encrypt and decrypt and is swept on every Lector startup. The X25519 secret key lives in the Windows Credential Manager, the macOS Keychain, or the Linux Secret Service, depending on platform. We never see it. We never sync it. Per-user storage, locked to your login.

You can export plaintext on demand: per document, not as a global toggle. Encrypt for archival, plain for sharing or backup. Recovery without Lector is a published path. Resero is a free, MIT-licensed companion tool that decrypts any library file using your stored key. Use it on a fresh machine, after uninstall, or for bulk export. It lives at useresero.app.

Methodology, threat model, and full source for the encryption module are in the substantiation paper at doi.org/10.5281/zenodo.19963302.

Lector is currently in development. Join the waitlist below and we will notify you when it ships. The launch price is $29, one-time, for up to three devices. No subscription. No telemetry. No account required. Lifetime updates for v1.x.