Quickstart

Get protected in 60 seconds

One command. No account. No configuration.

Node.js 18+Works with npm, pnpm, yarn, bunNo cloud account required
1

Install

Install once globally. Protects every project on your machine.

terminal
curl -fsSL https://pretense.ai/install.sh | sh

The package name on npm is pretense.

pnpmpnpm add -g pretense
yarnyarn global add pretense
bunbun add -g pretense
2

Initialize

Run once per project. Builds your mutation profile in .pretense/

~/your-project
pretense init
# Scans your codebase and builds .pretense/ mutation profile
# Takes ~3 seconds for a 100K-line codebase
# Nothing is sent to external servers during init

Creates .pretense/ in your project root.

3

Start the proxy

Keep this running in a terminal tab while you work.

terminal
pretense start
# Proxy running on http://localhost:9339
# All LLM API calls now route through Pretense
# Ctrl+C to stop

The proxy must be running before you send prompts. Add pretense start to your project package.json dev scripts to start it automatically.

4

Route your AI tool

Point your tool at localhost:9339. One change protects everything.

Claude Codebash
ANTHROPIC_BASE_URL=http://localhost:9339 claude

Set the env var once in your shell profile and every Claude Code session is automatically protected.

Cursortext
Settings → Models → Override Base URL
http://localhost:9339

Works with every model in Cursor: Claude, GPT-4o, Gemini. One setting protects all of them.

OpenAI SDKbash
OPENAI_BASE_URL=http://localhost:9339 node your-script.js

Drop-in replacement. No code changes — just set the env var before any OpenAI SDK call.

Any OpenAI-compatible toolpython
client = OpenAI(
    base_url="http://localhost:9339",
    api_key=os.environ["OPENAI_API_KEY"],
)

Pretense speaks both the Anthropic and OpenAI wire formats. Any tool that accepts a base URL works.

5

Verify it is working

Scan a file to see exactly what Pretense will protect.

terminal
pretense scan src/auth.ts
Output:
getUserToken->_fn4a2b(identifier)
API_SECRET_KEY->BLOCKED(secret)
DatabaseConnection->_cls8d3f(class)
secretKey->_v9x2c(variable)

Secrets are blocked before the request leaves your machine. Identifiers are mutated and restored transparently.

Free tier limits

The local CLI is always unlimited. Limits only apply to the cloud proxy service.

Proxy mutations
1,000 / month
Local CLI: unlimited
File scans
5,000 / month
Resets on the 1st
Audit log retention
7 days
Pro: 90 days
Team seats
3 seats
Pro: unlimited

Limits reset on the 1st of each month. View Pro plans ($29/seat/mo) when you need more.

What gets protected

Pretense is identifier-level. It mutates names, not logic. LLM output quality is preserved.

Token typeExampleWhat Pretense sendsStatus
Function namesgetUserToken_fn4a2bmutated
Class namesDatabaseConnection_cls8d3fmutated
Variable namessecretKey_v9x2cmutated
API keys / secretssk-ant-...BLOCKEDblocked
Comments// fetch user// fetch userpreserved
String literals"hello world""hello world"preserved

CLI quick reference

CommandDescription
pretense initScan codebase and build .pretense/ project profile
pretense startStart the proxy on localhost:9339
pretense scan <file>Scan a file for secrets and mutable identifiers
pretense mutate <file>Apply mutations and print the result
pretense reverse <file>Reverse mutations — restore original identifiers
pretense auditDisplay full mutation audit log (SOC2/HIPAA evidence)

Claude Code: MCP integration

Register the Pretense MCP server in your Claude Code settings for automatic pre-send scanning on every session.

.claude/settings.json
{
  "mcpServers": {
    "pretense": {
      "command": "pretense-mcp",
      "args": []
    }
  }
}

Next steps

You are protected. Here is where to go deeper.

Configuration
Mutation rules per repo, exclusions
Integrations
VS Code extension, GitHub Action, MCP server
How it works
Mutation algorithm deep-dive
Free vs Pro
When you need team features
Was this page helpful?