Your agents make thousands of decisions a day. Who’s authorizing them?
AI Guardian sits between your AI agents and the systems they touch. Every action is monitored, scored 0–100 for risk, checked against your policies and either allowed, flagged or blocked — before damage lands.
Enterprise agents need enterprise controls
AI agents now read your data, call your tools and act on your behalf. AI Guardian gives that autonomy a security perimeter — monitoring, control and accountability in one platform.
See everything
Every tool call, data access, network request and credential use your agents attempt is captured as a structured event — who, what, where, and how much.
Decide instantly
Each action gets a 0–100 risk score and a policy decision: allow, alert or block. Dangerous behaviour stops before it executes, not after it hits the news.
Prove it later
A complete audit trail with evidence, matched policies and analyst review status — the record your security team and your auditors will ask for.
Everything Guardian watches for
Nine event types, continuous scoring and a policy engine you control — the complete picture of what your AI agents actually do.
Real-time risk scoring
A 0–100 score for every action, built from the action type, data classification, destination, authorisation scope, timing and volume — with the reasoning shown factor by factor.
Policy enforcement
Express your rules as conditions on any event field. Policies decide allow, alert or block, and a matched policy sets the minimum severity the action is treated with.
Threat detection
Credential harvesting, data exfiltration, prompt injection, privilege escalation and shadow AI are detected and grouped into investigable threats mapped to MITRE tactics.
Complete audit trail
Immutable event records with evidence payloads, matched policies and risk factors. Analysts confirm threats or mark false positives, straight on the record.
Live event stream
Watch your agents operate in real time. Events stream into the console the moment Guardian scores them — no refresh, no delay.
API-first integration
One endpoint, one API key. Register an agent, stream its events and get scoring decisions back in the response — from any stack your agents run on.
Live in three steps
Register your agent
Add the agent in the console — its provider, autonomy level, data access and the capabilities you expect it to use.
Issue an API key
Create a scoped key with events:write. Point your agent at the Guardian ingest endpoint and stream every action it attempts.
Guardian stands guard
Each event is scored, matched against your policies and recorded. The decision comes back in the response so your agent acts only on what you allow.
Prompt filters aren’t enough
Other tools watch what your agent says. AI Guardian authorizes what it does — every tool call, checked against your policies before it fires.
| Capability | AI Guardian | Lakera | Portkey | Zenity |
|---|---|---|---|---|
| Real-time action blocking (before execution) | ||||
| API-call level access control | ||||
| Sub-10ms inline decision latency | ||||
| EU AI Act / NIST audit trail | ||||
| Free tier | ||||
| Developer SDK & API |
Based on publicly available product documentation as of October 2026.
Security teams sleep better with Guardian
Built for the teams every AI agent answers to.
“We rolled out 30 internal agents last year with no visibility into what they touched. AI Guardian gave us the audit trail our security team demanded — in an afternoon.”
“The risk scoring caught an agent pulling customer records off-hours in its first week. That one alert alone justified the subscription.”
“Policy-as-code for AI agents. Our compliance audit went from a three-week scramble to a one-hour export.”
Straightforward pricing
Every plan includes real-time scoring, policy enforcement and the full audit trail. Subscribe in minutes and cancel any time.
Free
For securing your first agent at no cost.
$0 /month
- 1 monitored agent
- 5,000 events / month
- 7-day event retention
- Core policies & risk scoring
- One-click policy packs
- Community support
Starter
For small teams shipping AI features.
$49 /month
- 10 monitored agents
- 50,000 events / month
- 30-day event retention
- All policy categories
- Inline guard API (authorize before execution)
- One-click policy packs
- Webhook alerts
Prices in USD. Cancel any time — annual plans save 25% and are billed once a year.
Frequently asked questions
How does AI Guardian see what my agents are doing?
Your agent streams each action to the Guardian ingest API — one authenticated call per event. Guardian scores it, evaluates it against your policies and returns the decision in the same response.
Can it actually stop an action?
Yes. Any policy can be set to block, and actions that cross the risk threshold are blocked automatically. A block decision comes back with the response, so your agent halts before executing anything dangerous.
Does it help with compliance and audits?
Every event is stored with its evidence, risk factors, matched policies and review status. Security analysts confirm threats or mark false positives on the record — a complete, exportable trail.
What counts as an event?
Tool calls, data access and transfers, network requests, credential access, privilege changes, model output, authentication and configuration changes — nine event types covering everything an agent can do.