TokenVeil is a self-hosted proxy that replaces sensitive data with tokens before your prompt reaches Claude, ChatGPT, Gemini or Mistral, then restores the real values in the response. Your team keeps the AI they want. The real data never crosses the network. Docker, 2-minute setup, 8 AI providers.
No reviews yetBe the first to leave a review for TokenVeil
Maker
📌
Hi PH 👋
I'm Marc, DevOps/Cloud architect. For the past two years, every client I worked with had the same conversation: "our people are pasting production logs and customer data into ChatGPT." And the only answer security teams had was "ban AI." Which never works. People just use their phone.
So I built the thing I wanted to have back then.
TokenVeil sits between your team and the AI. Before anything leaves your server, it replaces sensitive values (names, IPs, API keys, IBANs, customer refs) with typed tokens like . The AI reasons on the structure, answers with the tokens, and TokenVeil puts the real values back in memory. The real data never crosses the network boundary.
A few honest notes, because I'd rather you hear it from me:
It's alpha (0.1.0). It works, it's tested, but I'm not going to pretend it's battle-hardened.
The Community edition is source-available (Elastic License 2.0), not OSI open source. Read the LICENSE before you build on it. Regex engine only, no ML.
I measure 0% leakage, I don't guarantee it. 57 hand-annotated values, 3,283 fuzzed. The benchmark is public and reproducible on your own logs. Anyone promising you 100% is lying to you.
The NER engine (free-text names, orgs, locations) is the commercial edition. The regex one is free and covers the deterministic stuff.
Happy to answer anything, including the hard questions. What would you break first?
Report
Docker setup was genuinely painless, took about three minutes including the SSL stuff. Surprised how cleanly it handles the round-trip token swap without breaking chat formatting.
Report
Maker
@beratzgll Thanks Berat. The round-trip was genuinely the hard part. Detecting is the easy 80%; putting the real values back without mangling markdown, code blocks or the token positions in the model's answer is where most redaction tools give up (they just strip and never restore). Same value always maps to the same token, so relationships survive too. You sound like you'd break it in interesting ways, so if you find a case where the swap breaks formatting, that's exactly the bug report I want.
Report
docker setup was genuinely painless, had it routing through Claude in about three minutes. The fact that real customer data never leaves our infra is the kind of thing legal will actually get excited about.
Report
Maker
@sinan108042 Thanks Sinan, three minutes to Claude is exactly what I was going for. And you put your finger on the real selling point: the legal/compliance angle is usually what unblocks these deals internally. That's why there's a full RGPD dossier and a public reproducible benchmark, not just a "trust me". If your legal team wants to poke at it, the compliance page is on tokenveil.eu. Curious, what's the use case on your side?
Docker setup was genuinely painless, took about three minutes including the SSL stuff. Surprised how cleanly it handles the round-trip token swap without breaking chat formatting.
@beratzgll Thanks Berat. The round-trip was genuinely the hard part. Detecting is the easy 80%; putting the real values back without mangling markdown, code blocks or the token positions in the model's answer is where most redaction tools give up (they just strip and never restore). Same value always maps to the same token, so relationships survive too. You sound like you'd break it in interesting ways, so if you find a case where the swap breaks formatting, that's exactly the bug report I want.
docker setup was genuinely painless, had it routing through Claude in about three minutes. The fact that real customer data never leaves our infra is the kind of thing legal will actually get excited about.
@sinan108042 Thanks Sinan, three minutes to Claude is exactly what I was going for. And you put your finger on the real selling point: the legal/compliance angle is usually what unblocks these deals internally. That's why there's a full RGPD dossier and a public reproducible benchmark, not just a "trust me". If your legal team wants to poke at it, the compliance page is on tokenveil.eu. Curious, what's the use case on your side?