Stop rebuilding context
Create the reference once. A new conversation or a different AI no longer means finding and pasting the same three examples again.
Stop repeating “write like me.”
Build your voice once. Use the same version in your AI tools, then let Kaneme show what drifted before you publish.
- Free account, no card
- Your sources stay under your control
- Every verdict shows its limits
- Claude
- ChatGPT
- Grok
- Codex
The problem is not writing with AI.
It is rebuilding your identity in every tool, then correcting the result by instinct.
- 01
“Write like me.”
You have to find the examples again and explain the same preferences in every new tool.
- 02
“This is polished, but it is not me.”
The draft reads well while the author disappears from the sentences.
- 03
“I will rewrite the whole thing.”
The time saved by the model vanishes because no one shows where the voice drifted.
YourvoiceshouldremainrecognizableevenwhenanAIhelpsshapethesentences
One voice. Four controlled steps.
The reference survives the tool. Every change still belongs to the author.
- 01
Build
Choose writing you actually authored. Keep the sources, provenance and version together.
- 02
Carry
Call the same compact voice context from the assistant or workflow you already use.
- 03
Verify
Send the exact draft back to a measurement layer independent from the writing model.
- 04
Decide
Keep it, correct one passage, teach a preference, or refuse to let Kaneme learn from it.
Give every AI the same voice.
Paste one protected MCP address. Your selected voices, compact context and verification tools become available inside the client you already use.
Open connection guideThe model writes. Kaneme checks the distance.
The verifier is separate from the model that drafted the text. That separation makes the result useful across providers.
- 01
Your voice
A named version, human references, selected format and explicit permissions.
- 02
Your assistant
ChatGPT, Claude, Grok, Codex or another compatible MCP or REST client writes the draft.
- 03
Kaneme verdict
Q1 when supported, observed style gaps, bounded meaning checks and format reliability.
- 04
Your decision
The receipt records what was checked. Learning remains optional, reviewable and reversible.
What changes in real work
Kaneme keeps the reference portable, the review concrete and the learning reversible.
Keep one stable reference
Every use names the same voice revision. Your identity does not drift because today’s examples differ from yesterday’s.
Review evidence, not a feeling
The main verdict stays simple. The detail shows the measured distance, the remaining gaps and the limits of the result.
Let the voice evolve under your control
Preferences become proposals. You can inspect their references, edit them, exclude them, delete them or restore an earlier version.
A measured claim, with its boundary visible.
Production measurement is currently validated for French writing. The interface does not turn that evidence into an English engine claim.
Read the measurement boundary0,92
Published Q1 AUC
Can the served French engine separate this voice from AI writing in its measured domain?
0,82
Published Q2 AUC
Can it distinguish the selected author from another human writing in a comparable register?
AUC describes discrimination on an evaluation set. It is not an authorship certificate, a guarantee for every format or a promise of error free results. Kaneme can abstain when the text falls outside the measured conditions.
Your voice stays an asset you control.
- Human references keep their source and provenance.
- AI drafts never become human evidence.
- Connections and portable exports can be revoked.
- Every approved preference creates a restorable version.
Clear boundaries make the verdict useful.
- Not a tool for bypassing AI detectors.
- Not a claim that a voice match proves authorship.
- Not surveillance of someone who did not provide their corpus.
- Not an English verdict until the English engine is independently calibrated.
Questions before you build a voice
The limits are part of the product. They should be visible before you rely on a verdict.
Can Kaneme evaluate English writing?
Not yet. The product interface is available in English, while production voice verdicts remain calibrated for French texts. English measurement stays unavailable until a separate corpus, benchmark and calibration are complete.
Why is this different from persistent instructions in one AI?
Kaneme keeps the voice independent from the writing provider. The same named version can move between tools, and a separate engine can inspect the text produced by any of them.
Do I have to write inside Kaneme?
No. The native editor is a reference surface and a fallback. The strategic path is to use your voice through MCP or REST wherever you already work.
Is Kaneme an AI detector or a humanizer?
No. It does not help conceal AI use. Kaneme asks whether a text still carries a chosen voice and records AI origin when Kaneme generated the wording.
Can Kaneme prove that I wrote a text?
No. A voice match is not proof of authorship. Kaneme can verify which reference, version and text were measured, and it can explicitly abstain when the evidence is too weak.
What happens after Kaneme finds a drift?
You can keep the draft, correct a passage, propose a preference or learn nothing. Kaneme does not silently turn an AI output into a human reference.
Do I need a credit card to start?
No. You can create a free account and begin building your first voice without a card. Paid plans add continuity and capacity.
Which tools can connect today?
Claude, ChatGPT, Grok and Codex have verified Kaneme recipes. The same protected MCP address also works with other compatible clients, and the REST API covers custom workflows.
Start free. No card required.
Build your voice once. Keep it recognizable wherever AI writes.
Begin with your own French writing, inspect the reference, then call the same version from the tools you already use.
Build my voice