AGENTGUARD
DEMO
GitHub
SELF-HOSTED · POLICY-DRIVEN · MCP-NATIVE

A firewall for your AI agents

AgentGuard intercepts every tool call your agents make — allow, deny, or hold for human approval — explains the risk, maps it to the OWASP Agentic Top 10, and correlates it with what your EDR sees at the OS layer.

Allow / Deny / AskDeterministic policy on every call
Human-in-the-loopHold risky calls for approval
AI risk + OWASPScored, mapped to the Agentic Top 10
EDR correlationOS ground truth + bypass detection
runs in your browser · no installcontactmerveealpay@gmail.com·LinkedIn ↗
TOTAL
0
ALLOW
0
ASK
0
DENY
0
LIVE FEEDintercepted tool calls
0 eventsTAIL
⌁
Awaiting intercepts
Run the demo to stream live tool calls.
ASK QUEUE0
awaiting human review
✓
No calls awaiting review
Held tool calls will appear here.
UNCORRELATED EDR0
OS activity with no tool call
✓
No uncorrelated activity
Every OS event traces to a tool call.
AI ANALYSIS
Run an analysis to see results here.
AUDIT LOG
0 events
VerdictToolArgumentsOWASPRiskTime
no events