Skip to content
AIGuard · AI runtime security

Every prompt, tool call and answer, checked before it runs.

AIGuard checks every prompt, tool call and answer your AI apps and agents produce, blocks or holds the risky ones for a person, and records every decision.

AIGuard runtime security dashboard showing MCP scanner risk, tenant scope, connector health and top risks.
Runtime control plane
AIGuard AI inventory listing each agent, its model and its risk level.
AI inventory
AIGuard evidence catalog showing retention, legal hold and the compliance control each record maps to.
Audit evidence
ALLOWsupport-copilot · summarize ticket #4821BLOCKrag-search · injection in retrieved contextAUDITops-agent · mcp · db.execute DELETE held for approvalBLOCKhr-assistant · PII redacted from outputALLOWdev-copilot · mcp · github.read_file in scopeALLOWsales-assistant · CRM lookup, tenant verifiedAUDITfinance-agent · mcp · payments.refund above limitBLOCKkb-chatbot · system prompt override attempt
How it works

One check between your AI and your business.

Your apps don't change. AIGuard sits in the path every AI request already takes, and decides what is allowed to happen before it does.

  1. Your AI app or agent

    Sends a request or tries an action, as it does today.

  2. AIGuard checks it

    Against your policy: who is asking, what data is involved, what the action would do.

  3. Allowed, blocked or held

    Safe requests go straight through. Risky ones are blocked, or held until a person approves.

  4. Logged as evidence

    Every decision is recorded with its reason, ready for your auditors.

Where it runs
As a service, or inside your own cloud account
See a decision happen

Allow, block or hold. Before the action runs.

Your app or agent calls /api/ai/evaluate before each model or tool step. AIGuard checks it against your policy, returns a decision, and writes the evidence.

AIGuard · /api/ai/evaluatesimulated
Incoming request
support-copilot
Summarize ticket #4821 for the account owner.
GATE
Decision
ALLOW
In policy. Tenant scope verified.
19:17:10ALLOWsupport-copilot summarize ticket
19:17:10BLOCKrag-search injection in context
19:17:10AUDITops-agent db.execute DELETE
What it does

Four controls on every AI request.

01Stops attempts to trick your AI

Prompt and context inspection

Blocks instructions hidden in user messages or retrieved documents before they reach the model.

02Keeps agents to their job

MCP and tool authorization

Each agent can only take the actions it is allowed to. Risky ones wait for a person.

03Keeps sensitive data out of answers

Output filtering and redaction

Personal and confidential data is masked before it leaves your systems, which lowers GDPR and DPDP exposure.

04Records every decision

Audit evidence

Every allow, block or hold is logged with the policy that applied, ready for your review board.

Product demo

See runtime decisions, inventory and evidence.

The AIGuard console shows which MCP servers and agents are risky, which AI apps you run, and the evidence behind every decision.

Animated AIGuard demo showing runtime MCP risk, AI inventory and evidence screens.
Runtime decisions, inventory and evidence

Try AIGuard on one AI workflow for two weeks.

Week 1 only watches and records. In week 2 you review real decisions and the evidence behind them.

Request a 2-Week PilotSee all products
Other Ametriq products
For security operations teamsAgentic SecOps

Spectra

Turns hundreds of security alerts into a short list of cases, each with a named owner.
  • Related alerts grouped into one case
  • Every case has an owner
  • Nothing changes without a person approving it
See Spectra
For everyone at work who uses AIOn-device AI data protection · anti-scam

CyberShield

Stops company data leaking into AI tools, and scams reaching your people, on their own devices.
  • Redacts or blocks secrets pasted into 40+ AI tools
  • Flags phishing and fake payment requests in email and WhatsApp
  • Nothing is sent out to be inspected
See CyberShield