01
01 · Stop worrying
Stop worrying about AI privacy and security.
Shutapp redacts 300+ types of personal information worldwide, blocks all kinds of attacks, and restores the data when the AI responds.
Protect yourself, your users and ship AI with confidence. Keep sensitive data out of the model,
shut malicious attacks, and prove compliance without slowing down development.
SHUTAPP_API_KEY=sa_live_••••••••••••
Free. 2M Tokens / Month. No card.
3 Lines of code. Under 0.1 seconds latency.
Full privacy and security, without changing your existing LLM call.
drop into your code
# pip install shutapp
from shutapp import Shutapp
import os
s = Shutapp(api_key=os.environ["SHUTAPP_API_KEY"])
safe, session = s.scan(user_input) # before your LLM call
reply = call_your_llm(safe) # your existing call, unchanged
final = s.rescan(reply, session=session).text # what you return to the userShutapp live performance
Live shuts
—
shuts today
PII strips · injection / jailbreak · rescans
{ "receipt_id":"rcpt_pii_8f3a2b","decision":"shut","policy":"v0.5.1","proof":"EMAIL_1 + PHONE_1 → placeholders","signature":"Ed25519:7f3d9a41c80e6b2f54a11dc92468a3c19f0b72e5" }{ "receipt_id":"rcpt_pii_4e1027","decision":"shut","policy":"v0.5.1","proof":"CARD_1 blocked by policy","signature":"Ed25519:9a02c7e5b41d88f0372ac619ee5d4b7a03c8f17e" }{ "receipt_id":"rcpt_leak_91ac02","decision":"flag","policy":"v0.5.1","proof":"leak similarity 0.91","signature":"Ed25519:c41b7d3f921e4a0c8d55b7f60a19e29c74d30b3f" }{ "receipt_id":"rcpt_inj_b27f1d","decision":"shut","policy":"v0.5.1","proof":"instruction_override score 0.83","signature":"Ed25519:2e88f0a4d73c91b68af204d6c52a913ef8b70ca4" }{ "receipt_id":"rcpt_pii_aa55c9","decision":"pass","policy":"v0.5.1","proof":"no PII detected","signature":"Ed25519:5d10f4b77ac2680ef913dd4c2a7e60b19c45a9b7" }{ "receipt_id":"rcpt_resp_7d3e44","decision":"shut","policy":"v0.5.1","proof":"iframe stripped from reply","signature":"Ed25519:6b2c09f8de14a77c2e9305bb6d4a1c86f0279a09" }{ "receipt_id":"rcpt_pii_55b9a0","decision":"shut","policy":"v0.5.1","proof":"IBAN_1 + PERSON_2 + ADDRESS_1","signature":"Ed25519:14ef8bd2a90c771f4e62bb0c38d924a6f5e1d9d2" }{ "receipt_id":"rcpt_cost_1029ff","decision":"warn","policy":"v0.5.1","proof":"quota 91% — soft warn","signature":"Ed25519:8c446a3f01d7e9b25c80a44e31f62d7c9a0e566a" }{ "receipt_id":"rcpt_resp_3b8014","decision":"shut","policy":"v0.5.1","proof":"javascript: stripped from output","signature":"Ed25519:f0715b2e8c63a049bf7701dd9a5c418ee0b6145b" }{ "receipt_id":"rcpt_jb_a9c01e","decision":"shut","policy":"v0.5.1","proof":"jailbreak escalation pattern","signature":"Ed25519:3a9de8f70c12b64a91d0ef58c7a4b2e9f0139ce8" }{ "receipt_id":"rcpt_pii_8f3a2b","decision":"shut","policy":"v0.5.1","proof":"EMAIL_1 + PHONE_1 → placeholders","signature":"Ed25519:7f3d9a41c80e6b2f54a11dc92468a3c19f0b72e5" }{ "receipt_id":"rcpt_pii_4e1027","decision":"shut","policy":"v0.5.1","proof":"CARD_1 blocked by policy","signature":"Ed25519:9a02c7e5b41d88f0372ac619ee5d4b7a03c8f17e" }{ "receipt_id":"rcpt_leak_91ac02","decision":"flag","policy":"v0.5.1","proof":"leak similarity 0.91","signature":"Ed25519:c41b7d3f921e4a0c8d55b7f60a19e29c74d30b3f" }{ "receipt_id":"rcpt_inj_b27f1d","decision":"shut","policy":"v0.5.1","proof":"instruction_override score 0.83","signature":"Ed25519:2e88f0a4d73c91b68af204d6c52a913ef8b70ca4" }{ "receipt_id":"rcpt_pii_aa55c9","decision":"pass","policy":"v0.5.1","proof":"no PII detected","signature":"Ed25519:5d10f4b77ac2680ef913dd4c2a7e60b19c45a9b7" }{ "receipt_id":"rcpt_resp_7d3e44","decision":"shut","policy":"v0.5.1","proof":"iframe stripped from reply","signature":"Ed25519:6b2c09f8de14a77c2e9305bb6d4a1c86f0279a09" }{ "receipt_id":"rcpt_pii_55b9a0","decision":"shut","policy":"v0.5.1","proof":"IBAN_1 + PERSON_2 + ADDRESS_1","signature":"Ed25519:14ef8bd2a90c771f4e62bb0c38d924a6f5e1d9d2" }{ "receipt_id":"rcpt_cost_1029ff","decision":"warn","policy":"v0.5.1","proof":"quota 91% — soft warn","signature":"Ed25519:8c446a3f01d7e9b25c80a44e31f62d7c9a0e566a" }{ "receipt_id":"rcpt_resp_3b8014","decision":"shut","policy":"v0.5.1","proof":"javascript: stripped from output","signature":"Ed25519:f0715b2e8c63a049bf7701dd9a5c418ee0b6145b" }{ "receipt_id":"rcpt_jb_a9c01e","decision":"shut","policy":"v0.5.1","proof":"jailbreak escalation pattern","signature":"Ed25519:3a9de8f70c12b64a91d0ef58c7a4b2e9f0139ce8" }
Where Shutapp lands
Proof beats promises. Money backs it up.
AI security should not hide behind best effort. It should publish the numbers, put money behind the promise, and become a protection partner, not a disclaimer.
The top-right is the only quadrant worth owning: proven defense, real commitment, no hand-waving. Today, it's us. Alone.
Hover each circle to see the breakdown.
PromisedDefault ↑BundledSpecialists
shutapp
NV
SE
CF
CS
CP
LG
PK
EK
Proven Performance →
Commitment ↑
How it works. Secured end to end.
Every request is processed inside secure hardware enclaves. Your data remains completely private and is never stored or seen by anyone.
PII Redaction
Private values are replaced before the model call, then restored after the reply.Coverage~99.86%
User
1
The raw user prompt enters an isolated secure enclave for sensitive information redaction.
Originals are restored for your user, then all data is destroyed on delivery. Nothing is stored.
6Masking
2
Sensitive information is redacted. Only safe private text moves to the protection enclave.
The cleared reply moves back to the redaction enclave to restore the original values.
5Protection
3
Attacks gets blocked, only safe prompts goes to the LLM.
The LLM's reply re-enters the protection enclave for a harm-and-leak scan.
4LLM
Compliance proof
Every Shutapp decision produces a signed, cryptographic proof receipt. Your compliance evidence. Nothing personal is stored, only the signed evidence.
02
02 · Ship fast
Set and forget.
Three lines of code. Less than a tenth of a second and works with any AI model, continuously improving performance by our red team.
03
03 · Show proof
Compliance becomes visible, not promised.
Every check returns a small signed cryptographic receipt — a tamper-proof record of what happened. Your team and your auditors can verify it themselves, without ever needing to see the original data.
Simple pricing for every stage.
Start free. Upgrade when you need more capacity.
Free / Dev
$0/ month
1M input + 1M output tokens / month
Free
- All PII types redaction
- All attack patterns protection
- 1 seat
- Audit log of every request
Second most popular
Solo
$87/ month
15M input + 15M output tokens / month
$0.19 / 1M tokens
- All PII types redaction
- All attack patterns protection
- 1 seat
- Audit log of every request
Legal coverage
What we protect you from. In plain English.
What we stop, the risk if we don't, the article that names it, and the cost. Every citation resolves to a primary regulator source (EUR-Lex, ICO, eCFR, HHS, FTC, SEC, NIST, OWASP).
- GDPR
- EU AI Act
- UK GDPR
- HIPAA
- GLBA
- CCPA / CPRA
- OWASP LLM 2025
- NIST AI RMF
- ISO 42001
- SOC 2
Track 1
For the company
Six controls. Six article-level exposures.
C01Company exposure
Strip personal data from prompts before they leave your servers.
Without it · Personal data sent to the model lands in provider logs and downstream traces.
Cost · Up to 4% of global revenue (GDPR) · up to USD 1.9M / year per HIPAA violation type.
GDPR Art. 5(1)(c) · HIPAA 45 CFR 164.502(b)
C02Company exposure
Detect prompt injection and jailbreaks before they reach the model.
Without it · Untrusted input can override your instructions and trigger unauthorised actions or data disclosure.
Cost · Up to EUR 15M or 3% of global turnover (AI Act, operator obligations).
EU AI Act Art. 15(5) · OWASP LLM01:2025
C03Company exposure
Detect when the model echoes your system prompt back to a user.
Without it · Hidden instructions, pricing logic and tool definitions become public the moment they leak.
OWASP LLM07:2025 · GDPR Art. 5(1)(f)
C04Company exposure
Sanitise model output before your app renders it.
Without it · Unfiltered model output can carry executable markup that your frontend treats as trusted.
OWASP LLM05:2025
C05Company exposure
Cap tokens per API key on a rolling window.
Without it · A single misuse or runaway loop can exhaust budgets and trigger denial of service.
OWASP LLM10:2025
C06Company exposure
Sign every decision (Ed25519, RFC 8785) and verify it offline.
Without it · Without tamper-evident records you cannot prove what was processed, when, or under which policy.
GDPR Art. 5(2) · HIPAA 45 CFR 164.312(b) · NIST SP 800-53 AU-10
Track 2
For you, the developer, personally
Three places where the law puts a name on the indictment, not just a company logo. Receipts shift the burden of proof off you.
P01Personal exposure
In DE / IT / UK, mishandling personal data is a personal crime, not just a company fine.
Without it · Liability attaches to the individual; signed receipts evidence what was processed and how.
Cost · Up to 3 years prison (DE) · up to 3 years (IT) · unlimited fines on individuals (UK).
BDSG Sec. 42 (DE) · Codice Privacy Art. 167 (IT) · DPA 2018 ss. 170-173 (UK)
P02Personal exposure
HIPAA criminal liability reaches “an employee or other individual”, not only the entity.
Without it · Knowing disclosure of health data exposes the individual to fine and imprisonment.
Cost · Up to USD 250,000 and 10 years prison.
42 USC 1320d-6
P03Personal exposure
Solo developers shipping to EU users are personally the data controller.
Without it · The household exception does not cover a public commercial service (CJEU C-101/01).
Cost · Up to EUR 20M or 4% of your revenue, whichever is higher.
GDPR Art. 4(7) · Art. 2(2)(c)
Track 3
What we will not pretend to do
Where the line stops.
We do not run your DPIA, sign your DPA, file your 72-hour breach notice, or stand in for a HIPAA Business Associate Agreement. A gateway alone does not make you GDPR, HIPAA, or AI Act compliant. We make the evidence airtight; the legal paperwork is yours.
Informational only, not legal advice. Every cited article, section, and dollar figure is sourced in our internal research brief against tier-1 regulator and standards-body URLs only. Confirm against your own counsel before publishing or relying on any specific claim.
Line in the sand
TWO PROMISES · SIX HARD NOS.
Smaller perimeter, sharper language. Easier for security teams and the runtime owners who inherit this code.
- 01Reversible redaction.
Emails become [EMAIL_1], not █████. The model retains enough context.
- 02Signed receipts.
Ed25519 on every scan. Verify offline · auditors meet signed bytes.
- 03Honest injection handling.
Heuristic + ProtectAI scores land in your telemetry. Detection, not mythology.
Hard nos
- Promise to prevent prompt injection. Nobody honestly can.
- Solve hallucination at the scan layer. It's a different problem.
- Sit in the path of your model call. Your provider key is yours.
- Hide limits behind “Enterprise”. The OWASP breakdown above is canonical.
- Train on your prompts. No pipeline reads your payloads for modelling.
- Charge per seat. You buy tokens scanned.
Bring a dev
Both of you get +1M Input + 1M Output Tokens.
Stacks per invite. No cap. Credited the moment they generate a key.
You get
Input+1M
Output+1M
They get
Input+1M
Output+1M