aigate 0.1.1

AI Prompt Secret Scanner — local proxy that blocks secrets before they reach AI APIs

aigate

Local secret scanner that intercepts AI API calls and prevents credentials from leaking to LLMs.

Install

git clone https://github.com/jricramc/aigate.git
cd aigate
pip install -e .

Requires Python 3.11+ and jq.

Quick start

Claude Code (hooks — no proxy needed)

aigate install-hook

All prompts and tool calls are scanned automatically. Secrets are blocked before Claude sees them.

Any AI tool (proxy mode)

Terminal 1 — proxy:

aigate setup                         # one-time: installs CA cert (needs sudo)
aigate start --mode redact           # start the proxy

Terminal 2 — your AI tool:

source ~/.bashrc                     # load cert env vars (or open a new terminal)
export HTTPS_PROXY=http://127.0.0.1:8080
export HTTP_PROXY=http://127.0.0.1:8080
claude                               # or any other AI tool

aigate setup installs the mitmproxy CA cert into the system trust store and adds NODE_EXTRA_CA_CERTS to ~/.bashrc so Claude Code / Node.js trusts the proxy.

Scan a file directly

aigate scan .env
cat prompt.txt | aigate scan -

Modes

aigate start --mode block    # reject requests containing secrets (default)
aigate start --mode redact   # replace secrets with env var placeholders
aigate start --mode warn     # forward but log a warning
aigate start --mode audit    # forward silently, log only

Redact mode

Instead of blocking, redact mode rewrites the request before it reaches the AI:

1. Detects secrets in your prompt (AWS keys, API tokens, database URLs, private keys, etc.)
2. Replaces them with placeholders like [REDACTED_ANTHROPIC_API_KEY]
3. Saves the real credentials to a local .env file
4. Injects a system instruction telling the AI to use os.environ[] and load from .env
5. Forwards the sanitized request — the AI never sees the real credentials

The AI acknowledges the redaction, then writes secure code using environment variables automatically. Token prefixes are mapped to conventional env var names:

Token	Env var
sk-ant-*	ANTHROPIC_API_KEY
sk-*, sk-proj-*	OPENAI_API_KEY
ghp_*, github_pat_*	GITHUB_TOKEN
glpat-*	GITLAB_TOKEN
xoxb-*	SLACK_BOT_TOKEN
SG.*	SENDGRID_API_KEY
AKIA*	AWS_ACCESS_KEY_ID

Detection rules

• AWS keys — AKIA access key IDs
• API tokens — OpenAI, Anthropic, GitHub, GitLab, Slack, SendGrid, Square
• Database URLs — postgres, mysql, mongodb, redis, amqp, mssql with credentials
• Private keys — RSA, EC, DSA, OPENSSH, PGP
• Environment files — SECRET_KEY=value, DATABASE_URL=value, etc.
• GCP service accounts — JSON with type: service_account and private_key
• Tailscale keys — tskey-auth-*, tskey-api-*
• High-entropy secrets — password/token/secret fields with entropy > 3.5 bits

Logs

aigate logs          # last 20 entries
aigate logs -n 50    # last 50 entries
aigate logs -f       # live tail

Log file: ~/.aigate/scan.log

Docker

docker build -t aigate .
docker run --rm --entrypoint bash -it aigate
# inside the container, everything is pre-configured:
aigate start --mode redact &
curl -x http://127.0.0.1:8080 ...

Uninstall

aigate uninstall-hook   # remove Claude Code hooks