Now in development

Stop Detecting.
Start Proving.

AI detection is broken. False positives ruin lives. Provenance takes a different approach: instead of guessing what made something, it proves how it was made. Cryptographic proof of your creative process.

provenance-session — live capture
a3f8c1d2e4b6...timing_delta:42ms...dwell:89ms...flight:134ms...pause:2.1s...sha256:7e4a2b...block:#847291...verified b7e2f9a1c3d5...timing_delta:38ms...dwell:91ms...flight:128ms...pause:1.8s...sha256:9c1f3e...block:#847292...verified a3f8c1d2e4b6...timing_delta:42ms...dwell:89ms...flight:134ms...pause:2.1s...sha256:7e4a2b...block:#847291...verified b7e2f9a1c3d5...timing_delta:38ms...dwell:91ms...flight:128ms...pause:1.8s...sha256:9c1f3e...block:#847292...verified
The Problem

Detection is Broken

AI detectors don't work. They never did. They flag non-native English speakers. They accuse students who write well. They produce false positives that destroy academic careers. And every new model makes detection harder.

54%
False positive rate for non-native English writers on popular AI detection tools
$0
Accountability when a detector falsely accuses you. No appeals, no liability, no recourse.
?
Detectors can't tell you HOW something was made. They just guess. And guessing ruins lives.
"The question isn't 'Did AI write this?' The question is 'Can you prove you made it?'"
Three Classifications

Not Binary. Honest.

The world isn't "human vs. AI." That's a false binary that punishes legitimate use of tools. Provenance recognizes three classifications — and the third one changes everything.

Human-Created

Continuous typing patterns. Natural pauses for thought. Editing, revising, backtracking. A behavioral fingerprint that is unmistakably human.

Cryptographic proof

AI-Generated

No typing process recorded. Large blocks of text appear instantly. Paste-in patterns with no iterative creation process. The absence of proof is itself informative.

No process evidence

Human-Directed

The game-changer. You used AI as a tool — prompting, editing, revising, directing. Your creative process is still real. Your thinking is still there. Provenance proves it.

Process + direction verified

"Human-Directed" is the classification that education needs.
Using AI well is a skill. Provenance proves you have it.

How It Works

Four Steps to Proof

Provenance runs silently in the background while you create. When you're done, you have cryptographic proof of your entire creative process — without anyone seeing your work.

1

Create

Start Provenance and work normally. Write, code, design. The daemon captures keystroke timing — never content. Never keystrokes. Just the rhythm.

2

Fingerprint

On your machine, Provenance builds a statistical behavioral fingerprint — your unique creation pattern. Pauses, bursts, editing cycles, thinking time. All local.

3

Anchor

Only a SHA-256 hash — a meaningless string of characters — is anchored on-chain. Your fingerprint data never leaves your machine. Ever.

4

Prove

Present your Provenance Certificate. Timestamp, process classification, blockchain verification. Immutable proof that you made it, how you made it, and when.

Built Different

Sub-100KB binary. Zero dependencies. Written in Zig for maximum security and auditability. Open-source daemon — verify every line. Direct OS APIs, no frameworks, no supply chain risk.

Privacy First

Never Sees Your Content

Provenance is architecturally incapable of seeing what you type. Not "we promise not to look" — the code literally cannot capture content. This is verifiable, auditable, and open-source.

What We Capture

  • Time between keystrokes (milliseconds)
  • Key hold duration (dwell time)
  • Pause patterns (thinking time)
  • Edit/revision cycles

What We NEVER Capture

  • Which keys you press (no keycodes, no characters)
  • What you type (no content, no text, no passwords)
  • Screenshots, screen recording, or clipboard
  • Any raw data to any server (only a hash leaves)
Compile-time enforced: The Zig compiler prevents content capture at the type-system level. This isn't a policy — it's an architectural guarantee.
For Education

Built for the Classroom,
Not Against the Student

The current approach to AI in education is fear-based: detect and punish. But the students who learn to use AI well are the ones employers want. Provenance makes that skill visible and verifiable.

The Old Way

Run student work through a detector. Get a percentage. Accuse the student. No proof. No appeal. Relationship destroyed.

The Hypocrisy

Faculty use AI to draft emails, prep lectures, and write grants — then punish students for the same behavior. The double standard is unsustainable.

The Provenance Way

Students prove their process — human-created or human-directed. Both are legitimate. Both demonstrate real learning. Both earn the grade.

The Future

AI literacy is the new digital literacy. Students who learn to prompt, direct, edit, and refine are building the skills that matter. Provenance makes that visible.

"How many of us used AI to prep for this panel? Would we fail a student for doing the same thing?"

Provenance eliminates the hypocrisy by making process visible — for everyone.

Early Access

Be First to Prove It

Provenance is in active development. Join the waitlist to get early access when we launch. Students, creators, educators — you're the people we're building this for.

No spam. No selling your email. We'll notify you when early access opens.

$24.99/year
Open source daemon
All platforms
Never sees content