Runtime guardrails for AI agents, apps and APIs. MoorAI reviews prompts and tool calls before they run, coaches users in the moment, and gives security teams content-free visibility: the checks run where the agent runs, so sensitive content is never sent anywhere to be checked.
Open source (MIT) and free — no account, no credit card, nothing to stand up. The agent guards the device on its own; the console is optional: self-host it free, with no user limit, or start a free trial of the hosted console. Also available for Windows.
Each message is reviewed against 67 AI-security threats + content policy before it reaches the agent. Block, alert, or allow — your call, per tenant.
Detection runs on the device. Only redacted metadata (no raw content) is reported to the portal — so the guard never becomes the leak.
A security dashboard with per-tenant policy, device inventory, and a live detection feed — so you can see and tune AI risk across the org.
The same on-device review, with the exact categories you enforce shown below. Every category is set per device to one of four actions.
67 AI-security threats across these policy areas, plus on-device DLP for secrets and personal data:
Self-host free, no user limit · Hosted: free trial, then contact us for pricing
Six content categories applied to what's sent to — and returned by — the AI:
Always free
Protection coverage by language today — we're explicit about where it's strong and where it's growing:
Emails, national IDs, payment cards, IBANs, API keys, private keys, JWTs, phone numbers, code blocks & URLs are matched by pattern — so they're caught regardless of the language you write in.
Parental-control content (explicit, violence, self-harm, drugs, hate, profanity) now matches English & Hebrew. Prompt-injection and social-engineering (BEC) threat detection are keyword-based and English today.