Prompt-injection defense
Gate scans every prompt before it reaches the model and blocks injection attempts, including indirect attacks hidden in tool output. See the prompt-injection scanner.
Gate documentation
Gate is a drop-in security and audit layer for the AI tools and models your team already uses. It sits between your tools and the model providers and screens every request in both directions: blocking prompt-injection attacks, redacting secrets and personal data, and sealing every decision to a tamper-evident audit trail. It trims token cost along the way. Your tools keep working the way they do today, with no change to your code.
Start here
Run agent and desktop tools Route Claude Code, ChatGPT, and the other tools you already run through Gate from a menu-bar app. No code.
Set up manually Point Claude Code, Codex, or your own SDK at Gate by changing one setting. No app to install.
New to Gate? How Gate works, and the two choices you make when setting up: how you connect and how you pay.
What Gate does
Secret and PII redaction
Credentials and personal data are caught and stripped from responses before they reach your tools. See Security.
Tamper-evident audit trail
Every prompt, reply, and rule decision is written to a verifiable log you can inspect later. See the audit trail.
Lower token cost
Compression and caching cut what you spend, with the same model output.
One endpoint, every model
Reach providers like Anthropic, OpenAI, Bedrock, and Vertex through a single API.
Explore the docs
Plans and pricing What the Free and Pro plans include, and how billing works.
Security overview How Gate protects your requests, data, and keys.
Dashboard tour Find your keys, usage, audit log, and settings.