Your machine
Real code + secrets
Secrets never leave unmasked
Stop Leaking Secrets to AI
You can suspend a model. You can't recall a disclosure.
Your engineers paste credentials, API keys and customer data into AI tools every day. Pretense swaps them for stand-ins before they leave the laptop.
Deploy in 30 seconds
No credit card required
Works with your AI tools
Works with your existing AI tools
Works with any tool that accepts a custom API base URL. Verified end to end with Claude Code, Gemini and Cursor.
The Blind Spot
Samsung engineers leaked proprietary code to ChatGPT three times in under a month (2023).
Nobody in that building knew. Not security. Not legal.
This is the part nobody says out loud. Your developers are already using AI every single day, and the security tools you bought were built for a different problem. They watch email and file uploads. They were never built for a senior engineer opening Claude at 2am to fix a release.
The industry spent two decades building walls around the front door. The code walked out through a browser tab.
unmonitored · ai_traffic.log
[09:14:02] → OUT ChatGPT
api_key=sk-prod-7f3a…
[11:42:55] → OUT Claude
patient_ssn=412-55-0199
[14:07:31] → OUT Copilot
// full auth/service.py
[23:58:10] → OUT ChatGPT
db_pass=Pr0d!9x2k
4 exposures · 0 logged · 0 alerts
What It Costs
And it is already costing companies money
This is not a future risk. IBM put numbers on it this year.
+$670K
added to a breach when staff use AI tools nobody approved
IBM Cost of a Data Breach 2025
1 in 5
breaches now involve one of those unapproved AI tools
IBM Cost of a Data Breach 2025
97%
of AI-related breaches happened with no access controls in place
IBM Cost of a Data Breach 2025
How It Works
So we built a firewall for the thing they actually use
Pretense sits between your developer and the AI, like a proxy on the machine. When code goes out, it finds the sensitive parts and swaps them for stand-ins. The AI does its work on the safe version. When the answer comes back, Pretense puts your code names back in place. Secrets and PII stay as stand-ins and are never put back.
The developer writes code the way they always have. The AI answers the way it always has. The passwords, the patient records, the real names in your code: none of them ever leave the laptop.
Mutate
Pretense
Swaps secrets for stand-ins
Safe prompt
The AI
Sees only stand-ins
← code names restored in the reply; secrets and PII stay as stand-ins
Security stops being a thing you bolt on after the fact. It becomes part of the conversation with the AI itself.
Features
Everything You Need To Stop AI Leaks
Track daily code protection activity trends
Visualize Code Protection trends over time and understand how AI interactions impact your codebase.
Real-time Code Protection trends and growth- Identify spikes and unusual activity
- Filter by time range and usage patterns
10:55:10 | [SUCCESS] | API key mutated |
|---|---|---|
API key mutated | ||
frontend-web by Emily Tran | src/users/service.py | |
src/users/service.py | ||
10:54:22 | [BLOCKED] | Request blocked: unrecognized secret |
|---|---|---|
Request blocked: unrecognized secret | ||
backend-service by John Doe | src/db/connect.py | |
src/db/connect.py | ||
10:53:55 | [PROTECTED] | SSN mutated |
|---|---|---|
SSN mutated | ||
frontend-web by Emily Tran | src/users/service.py | |
src/users/service.py | ||
10:52:46 | [SUCCESS] | Email address mutated |
|---|---|---|
Email address mutated | ||
frontend-web by Emily Tran | src/users/service.py | |
src/users/service.py | ||
See activity as it happens in real time
Monitor live Code Protection across your system with instant updates and status visibility.
- Real-time stream of Code Protection activity
- Status indicators (Success, Blocked, Mutated)
- Track user, repo, and file-level activity
Total Code Protection | |
|---|---|
1542 | vs Apr 10- Apr 18 |
Top user | |
|---|---|
 Jimmy Malhan | 28,164 Protects |
Most active repo | |
|---|---|
| 5,593 Protects |
pretenseTotal Code Protection | |
|---|---|
Scan Proxy | 1120 (73%) 422 (27%) |
Get insights at a glance across your system
Understand key metrics, top contributors, and overall Code Protection activity in one place with better clarity.
- Total Code Protection and growth trends over time
- Top users and most active repositories overall
- Action breakdown & usage analytics across teams
Mutation, not Redaction
Hiding the code shouldn't make the AI worse
Other tools redact or block the sensitive parts. That deletes the context the AI needs, and the answers get worse with every prompt. Pretense swaps instead of deletes, so the AI never knows anything is missing.
Redact or block — the old way
The AI gets blanks
The sensitive fields are stripped out before the prompt is sent. The model loses the thread and starts guessing.
# what the AI receives
patient = [REDACTED]
ssn =
record = [REMOVED]
Over a few prompts the accuracy keeps slipping.
Mutation — the Pretense way
The AI gets working code
The values are swapped for realistic stand-ins on your machine. The AI works normally. Pretense reverts them in the response.
# what the AI receives
patient = "alpha_record"
ssn = "BETA-4471-XX"
record = "mrn_beta_009"
Full accuracy, same speed, and the values that matter never leave the machine.
GETTING STARTED
You can have this running before your coffee is cold
No procurement process. No rebuilding your stack. Four steps and you are protected.
Install the CLI
One command. Works on macOS and Linux.
$ curl -fsSL https://www.pretense.ai/install.sh | sh
Create your organization
Name your team so activity and access stay in one place.
Generate an API key
This connects Pretense to the AI tools you already use.
Start coding
Sensitive data gets swapped automatically before it reaches any model.
That is the whole setup. From install to first protected prompt is about 30 seconds.
security compared
Trusted protection for modern AI workflows
Feature | Pretense | Traditional DLP | No Protection |
|---|---|---|---|
Code Protection (preserves AI context) |
|
|
|
Deploys in 30 seconds |
|
| - |
Works offline (local-first) |
|
|
|
Multi-provider (Claude + GPT + Gemini) |
|
| - |
Audit trail with SOC 2 / HIPAA policy presets |
|
|
|
Starting at $0/month |
|
|
|
PRICING
Choose the plan that fits your workflow
MonthlyYearly20% OFF
Free
Perfect for getting started with code protection
$0 /month
Get Started Free
- 10,000 code protections / 7 days
- Up to 3 seats
- 7-day log history
100,000 code protections / 7 days- More than 3 seats
- Usage CSV export
- Priority support
Startup
Built for growing teams that need more power and flexibility
$29 /seat/month
Upgrade to Startup
- 100,000 code protections / 7 days
- 1–100 seats
- 30-day log history
- Priority email support
- Usage dashboard & CSV export
- Dedicated support & CSM
Enterprise
For large organizations with advanced security needs.
$99 /seat/month
Contact sales- Everything in Startup
- 50 seats minimum, up to 100
- Dedicated support & CSM
- SLA & uptime guarantee
- SAML / SSO (Coming Soon)
Secrets never leave your system unmasked
Built for secure workflows
Works with your existing tools
Cancel anytime seamlessly
FAQS
Got questions? We've got answers.
Here's everything you need to know before getting started.
Pretense sits between your developers and the AI tools they use, like a proxy on the machine. Before any code or prompt reaches an LLM, it finds the sensitive parts and swaps them for realistic stand-ins. The AI works on the safe version, and when the response comes back, Pretense restores your code names. Secrets and PII stay as stand-ins and are never put back into the reply. Your real secrets and code names never leave the machine.
Start now
Start where you are
However you like to start, there's a way in.
Ready to evaluate
Talk to us
Walk through compliance coverage and audit trails with the team.
Want to try it
Start free
Install it on your own machine in 30 seconds. No credit card.
Just looking
See it work
Watch a 2-minute walkthrough of mutation in a live codebase.