Connecting…

YOUR SECURITY, AT A GLANCE

A little more peace of mind.

Visibility into what your agents read. Control over what gets through.

Requests inspected—Latest 500 requests
Requests blocked—Policy, content & network checks
Awaiting review—Pending write approvals
Observed agents—In the loaded request window

Request activity

Traffic through your security gateway

Last 24 hours
24 hours ago12 hours agoNow
AllowedBlockedHeld for review
◇

DEFENSE IN DEPTH

A safer path to the web.

Every routed request is checked against your agent’s policy and screened for known injection patterns.

  • Prompt-injection screening
  • Domain & network checks
  • Write-action approval gate
Prototype protection reduces risk. It does not eliminate prompt injection.

HUMANS IN THE LOOP

Your agents asked. You decide.

Write actions are held here until you approve or deny them.

0 pending
AGENT / REQUESTMETHODWAITINGACTIONS
✓

Nothing waiting on you.

When an agent attempts a write action, it will appear here for your review.

Recent activity

Your agent requests will appear here.

AGENT / DESTINATIONDECISIONSCORELATENCYTIME
◎

Ready when your agents are.

Route a request through /proxy to see its decision here.
Sanitizer lab previews are not recorded as proxy traffic.

IDENTITY MEETS CONTROL

Your agents. Their boundaries.

Individual keys, independent policies, and access you can switch off.

Connecting to agent management…

AGENTACCESSAPI KEYCREATEDACTIONS

EXPLORE YOUR DEFENSES

Meet your sanitizer.

Preview how AgentAIGuard handles untrusted content. No external requests are made.

Local preview
Sanitized outputNot scanned
Your scan results will appear here.

Known patterns are detected using rules, not a trained classifier.

Connect to your workspace

If this instance has an ADMIN_TOKEN, enter it to view activity. The token stays in memory and is cleared when you reload.

Review write action

Approving releases this request for the agent to retry. Verify the destination and payload before allowing it.

Legal notice

Operator
Contact
Use the “Contact the operator” form in the footer; messages go directly to the operator.
Security disclaimer
AgentAIGuard reduces the risk of prompt injection and uncontrolled agent actions. It does not — and cannot — eliminate them. Detection is rule-based: injections using phrasing, encodings, or languages outside the known patterns can still reach the agent. Per-IP rate limits do not stop distributed attacks. The human approval gate for write actions is the fallback control — review held actions before approving. Do not rely on this service as your only control for untrusted content.
Data notice
This service logs every proxied request: agent identity, destination URL, method, decision, detection score, and payload excerpts for held actions. Approval queue contents may include request bodies. Do not send credentials or personal data through the proxy.
License
AgentAIGuard is licensed under the Business Source License 1.1 — free for internal production use; offering it as a hosted service to third parties requires a commercial license. Provided without warranty.

Contact the operator

Your message goes to the operator of this AgentAIGuard instance. They will reply to the email you provide.

Request details

Create an agent

A unique key will be generated. Save it somewhere secure; it cannot be retrieved later.

Save your agent’s key.

This key is shown only once. Store it in your agent’s secret manager. Closing this dialog clears it from the page.

Connect your agent

Send this key as X-AgentGuard-Token or Authorization: Bearer <key> on requests to /proxy?url=<encoded destination>. No X-Agent-Id is needed. Use HTTPS outside local development.