A single gateway to all LLMs with data masking and prompt-injection protection. Employees use AI legally — confidential data never leaves the perimeter.
FAQ
Does data leave our perimeter?
With On-Premise deployment the gateway sits inside your perimeter, and we as the vendor do not see your traffic. In sovereign mode everything stays inside on local models; with foreign models only de-identified data leaves.
Do we need to rewrite our systems to connect?
No. Applications reach AI models through our gateway address instead of calling the model directly. All traffic automatically passes the checks.
Is a GPU required?
No. Detection runs on lightweight ML models and regexes on CPU.
What data is masked?
Legally protected personal data by default; industry and corporate confidential data is configured per company. Masking applies to any model, including GigaChat, YandexGPT and local ones.
Can it integrate with our SIEM?
Yes. Logs of detections and traffic export to your SIEM as JSON via API. If there is no SIEM, Gatewarden collects, stores logs and generates alerts itself.
Do you protect against secret leaks and dangerous AI-agent actions?
Yes. The gateway blocks sending secrets to AI (API keys, tokens, JWT, private keys, ~1600 formats), and destructive SQL from AI agents is blocked before execution (OWASP LLM06).
How does this help with compliance?
The product covers 152-FZ and 187-FZ, deploys in a closed perimeter and is designed for FSTEC order №117. Every triggered check is recorded in the audit log.