How Solus works

Stop guessing whether AI wrote it.
Prove that a person did.

AI detectors are broken. They flag honest students, miss actual cheating, and turn every accusation into a fight neither side can win — because they're trying to answer the wrong question. Solus flips the problem. It's a writing environment AI can't use in the first place, where every keystroke, paste, and input is recorded as it happens. Authorship becomes provable, not probable.

Who this is for

Every accusation is a guess. That's the problem.

If any of the below sounds like your week, Solus was built for you.

Teachers & professors

You can't grade with confidence. Every AI-detector flag turns into an argument you can't win.

Every submission carries a report showing exactly how the document was written. Not an accusation. Evidence.

Students

Your own writing gets flagged as AI. You have no way to prove otherwise.

Write in Solus and every document ships with proof it's yours — keystroke by keystroke. Not a probability. Verifiable.

Schools & district admins

You're stuck between two failure modes — miss real cheating or punish innocent students.

Institutional verification of authorship. No black-box classifier. No false positives. Just evidence.

Journalists, lawyers, consultants

How do you prove to a skeptical client that your draft is really yours?

Send them the .solus file. They drop it into the public verifier. Deterministic proof of authorship — no account, nothing uploaded.

The switch

Verify writing at the source. Don't classify it after.

AI detectors ask: "was this text generated by a machine?" — a question no one can answer for sure. Solus asks a different question — "how was this text actually written?" — and captures the answer as it happens.

Old way · AI detectors

Try to detect AI in the finished document

  1. 1Student writes wherever — Word, Docs, notes app.
  2. 2Teacher pastes essay into an AI detector.
  3. 3Detector guesses: "87% likely AI".
  4. 4Teacher confronts student. Student denies.
  5. 5Nobody can prove anything. Trust collapses.
Solus way · Verified at the source

Capture the writing process, then verify it

  1. 1Student writes in Solus — an environment AI can't use.
  2. 2Every keystroke, paste, and input recorded live.
  3. 3Submission ships as a signed .solus file.
  4. 4Teacher opens it — gets Verified Human™ or the exact failure reason.
  5. 5Deterministic. Not a probability. No arguments.
How verification works

29 checks. Two axes. One deterministic verdict.

Nothing here is a probability. Every check either passes, fails, or explicitly skips — and you can see the result of each one individually. The Verified Human™ badge is only awarded when both axes come back clean.

Axis A · 19 checks Provenance Integrity

The archive is intact. The certificate hasn't been altered. The keystroke ledger still reproduces its commitment. Plain-English: nobody has touched this file since it was signed.

Any single failure → verdict is tampered.

Axis B · 10 checks Human-Origin Support

The typing pattern in this document matches the writer's own historical rhythm. Composition beats transcription. No synthetic input. Plain-English: this person actually typed this, and their fingers know it.

Hard fail → unsupported. Partial → limited.

Every check is visible in the verification report — you can see exactly which one passed, which failed, and why. See all 29 checks with plain-English detail →
Three things AI detectors can't do

The attacks that break detectors. Solus catches all three.

Attack 1

Copy-paste from ChatGPT

Sometimes flags it, sometimes doesn't. Depends on how “AI-like” the wording is.
Paste events are cordoned off with visible policy tiers. Every paste is in the record — sealed, attributed, or blocked, by institution setting. No hidden clipboard bypasses.
Attack 2

Type from a phone or second screen

Text looks like ordinary writing. No hope of detection.
Composition and transcription look identical in the finished text, but leave completely different typing rhythms. A five-signal composite catches the difference — pause locations, attention continuity, dynamics, non-linearity, word-timing. How →
Attack 3

Auto-typer or macro injection

Text is real writing. Nothing to detect.
Programmatic input is caught by the writing environment itself — a native keystroke-origin hook on Windows, script-event blocking on web. Any injected input lands in the “rejected” bucket, never in accepted content.
What Solus is not

Not surveillance. Not a black box. Not a classifier.

The premise sounds invasive on paper. It isn't. Here's what actually happens.

Your content is encrypted end-to-end.

Document text is sealed with AES-256-GCM and wrapped for the intended recipient (usually your teacher). Solus's servers never hold the decryption key. Nothing you write is legible to us — ever.

Verification runs entirely in your browser.

When someone drops a .solus file into the public verifier, the file is not uploaded anywhere. All 29 checks run locally using the Web Crypto API. You can disconnect from the internet before dropping a file and it still works.

Every check is individually visible.

No opaque "authenticity score." The report shows 29 separate pass/fail results with plain-English explanations. If something fails, you know exactly which check and why — not a probability, not a mystery.

Only pattern data is analyzed — not what you wrote.

The behavioral checks look at typing dynamics (dwell, flight, burst rhythm) and event ordering. They don't read your essay. Teachers see the writing when they open the doc; the verifier can score behavior without ever decrypting content.

Earning the badge

How a writer becomes Verified Human™

1

Write ~3 normal sessions.

Just do your work. Solus learns your typing rhythm silently in the background. About 8,000 keystrokes across three or more days is enough.

2

Enrollment completes.

Your baseline is sealed. Now every submission runs both verification axes against your own history.

3

Both axes green → the badge.

Verified Human™ is awarded when provenance is intact and the typing pattern matches your baseline.

Your first few documents show “Enrolling.” That's normal — not a failure. Cryptographic provenance is still verified; the behavioral axis just needs runway. For classroom deployments, teachers see this state explicitly and grade accordingly.
Where next

Drill into your part of the story.

For teachers

Grade with evidence

For schools

Institutional verification

For students

Prove your work is yours

For professionals

Send provable drafts

Ready to see it work?

Try the public verifier with any .solus file, start writing free, or spin up a teacher account with a 60-day pilot — no card, no approval, no wait. Teachers can also preview the institution portal in a self-serve sandbox.