Security and data

Your data never has to leave your environment.

A tool that tests agents for data leaks shouldn't create new ones. Olyptic is built to run where your data already lives.

Self-hosted or UK-hosted

Run the whole platform inside your own cloud or data centre. If you'd rather we host it, it runs in the UK, in an environment dedicated to you.

Replicas, not your live systems

Agents are tested against seeded replicas of their tools: inbox, files, CRM, payments. Tests never send a real email or move real money.

A judge that can stay local

Where an AI judge is needed, it can be a model running on your own hardware. No prompts or traces go to an outside model provider unless you choose to.

A locked-down test sandbox

During a test, the agent can only reach its replicas and its own model endpoint. Requests to cloud metadata services and internal addresses are blocked.

Secrets stay secret

Credentials for your agent are encrypted at rest and never written into scenario files, traces or reports.

Evidence you can check

Every report records the scenario versions, model and settings used. Signed evidence packs, with hashes of every trace, are on our roadmap.

Production monitoring: opt-in, and still to come

Watching agents in production is on our roadmap. When it arrives, it will be opt-in, read-only, and limited to the trace data you choose to send. Until then, everything we do happens on replicas.

Questions from your security team? Email security@olyptic.com.