Not prompts · Not guardrails

Your sensitive data never reaches a human — or an AI.
And you can prove it.

Not a prompt. Not a guardrail. A cryptographic guarantee, checked on every request — not a model you asked to be careful.

See it decide

A request arrives. The engine hands back only the authorized fields plus a signed receipt — everything else stays sealed.

The decision below is computed live on the real engine, proof hash and all. Hit "Run again".

POST /api/access/compute

subject"agent:claude"

resource"insurance-claim/4820"

intent"summarize"

Claim statusRegion🔒 National ID🔒 Claim amount

200 OK

computing on the real engine…

See it on the Fabric page →Verify a receipt yourself →

One document, end to end

Watch one real document get classified, its sensitive fields sealed, and only the allowed ones reach a language model that answers.

Every step runs the real engine. Tap a step to open its explanation.

A sample scan, then yours

See a folder scanned in seconds — then run the exact same scan on your own folder. Nothing uploads.

Scan a folder →

Connect your AI

One line wires any MCP agent — Claude included — through the engine, so every data request it makes is decided and proven.

$claude mcp add nodata -- npx @nodatachat/mcp --api-key sk_live_YOUR_KEY

Get a free key →

And it holds at scale. The exact same proof over one document or a hundred million — one policy recomputed on every request, never a permission cached, so a revocation takes effect at once.

Want it on your own problem? · Go deeper

Pick the problem you know

Pick a scenario and see exactly what the Capsule does with it. Without ever seeing your data, without holding the keys.

Rocky

Capsule guide

Let's see what a model gets when it tries to read sealed data.

Watch the mechanism play. I'll translate it at the end.

customer_record.jsonREADY

FieldSeverity / State

ssnCRITICAL

emailHIGH

first_nameLOW

plan_tierLOW

Policynot yet applied

Every action is signed with proof · Verify

LegendCRITICALHIGHMEDIUMLOW🔒 sealed · key not derivedsigned · proof

🔎 Scan your own folder →The same content-blind engine, in your browser, nothing uploads.

Area 3 · Q&A

🔒See 1 more example to unlock Q&A. First you see the mechanism, then you ask.

We secure your data, without ever seeing it, without holding the keys, even after it leaves your organization, with cryptographic proof for every disclosure and touch.

Verify a pack Talk to us →

Start free

Not prompts · Not guardrails

Your sensitive data never reaches a human — or an AI.
And you can prove it.

Not a prompt. Not a guardrail. A cryptographic guarantee, checked on every request — not a model you asked to be careful.

See it decide

A request arrives. The engine hands back only the authorized fields plus a signed receipt — everything else stays sealed.

The decision below is computed live on the real engine, proof hash and all. Hit "Run again".

POST /api/access/compute

subject"agent:claude"

resource"insurance-claim/4820"

intent"summarize"

Claim statusRegion🔒 National ID🔒 Claim amount

200 OK

computing on the real engine…

See it on the Fabric page →Verify a receipt yourself →

One document, end to end

Watch one real document get classified, its sensitive fields sealed, and only the allowed ones reach a language model that answers.

Every step runs the real engine. Tap a step to open its explanation.

A sample scan, then yours

See a folder scanned in seconds — then run the exact same scan on your own folder. Nothing uploads.

Scan a folder →

Connect your AI

One line wires any MCP agent — Claude included — through the engine, so every data request it makes is decided and proven.

$claude mcp add nodata -- npx @nodatachat/mcp --api-key sk_live_YOUR_KEY

Get a free key →

And it holds at scale. The exact same proof over one document or a hundred million — one policy recomputed on every request, never a permission cached, so a revocation takes effect at once.

Want it on your own problem? · Go deeper

Pick the problem you know

Pick a scenario and see exactly what the Capsule does with it. Without ever seeing your data, without holding the keys.

Rocky

Capsule guide

Let's see what a model gets when it tries to read sealed data.

Watch the mechanism play. I'll translate it at the end.

customer_record.jsonREADY

FieldSeverity / State

ssnCRITICAL

emailHIGH

first_nameLOW

plan_tierLOW

Policynot yet applied

Every action is signed with proof · Verify

LegendCRITICALHIGHMEDIUMLOW🔒 sealed · key not derivedsigned · proof

🔎 Scan your own folder →The same content-blind engine, in your browser, nothing uploads.

Area 3 · Q&A

🔒See 1 more example to unlock Q&A. First you see the mechanism, then you ask.

We secure your data, without ever seeing it, without holding the keys, even after it leaves your organization, with cryptographic proof for every disclosure and touch.

Verify a pack Talk to us →

Your sensitive data never reaches a human — or an AI.And you can prove it.

Pick the problem you know

Your sensitive data never reaches a human — or an AI.And you can prove it.

Pick the problem you know

Your sensitive data never reaches a human — or an AI.
And you can prove it.

Your sensitive data never reaches a human — or an AI.
And you can prove it.