Secure API keys
Project API keys are hashed at rest and shown only once on creation.
AgentGuardSecure every AI agent before it acts
Inline AI guardrails
Keep every AI workflow inside policy.
AgentGuard blocks prompt injection, redacts sensitive data, validates tool calls, and records audit-ready evidence before risky AI actions reach production systems.
Prompt injection defense
Detect role override, jailbreak, and system prompt extraction attempts before model calls.
One-time API keys
Hash project keys at rest, reveal secrets once, and track traffic by key prefix.
Resolution workflow
Review open findings, exceptions, retests, and remediation audit trails.
Access inventory
Track actors, devices, MCP servers, tools, and high-risk actions across AI traffic.
Overview > Security dashboard
Search incidents, policies...
Security overview
Last updated: 2026-06-26 14:32 UTC - 8 models monitored
Requests blocked
1,621
Today up +12.4%
Active policies
5
Today up +12.4%
Models monitored
8
Today up +12.4%
Compliance score
97.3%
Today up +12.4%
Request traffic - today
Total requests vs. blocked
- Requests - Blocked
Violations by type
Last 24 hours
PII
Injection
Jailbreak
Bias
Recent incidents
Incident IDModelPolicySeverityStatus
INC-4821GPT-4oPII DetectionCriticalBlocked
INC-4820Claude 3.5Prompt InjectionHighBlocked
INC-4819Llama 3.1Jailbreak AttemptHighBlocked
Security platform
Govern prompts, responses, tools, and fixes.
Add policy enforcement without rewriting your AI stack. AgentGuard centralizes scanner coverage, access visibility, logs, reports, and remediation in one workspace.
Tenant isolation
Supabase RLS keeps organizations, projects, logs, and reports isolated.
Evidence-ready logs
Every check records sanitized reasons, severity, risk score, and decision.
Red-team validation
Run adversarial test cases to catch guardrail gaps before release.
Live gateway console
Every AI action moves through policy.
Inspect, score, enforce, and resolve traffic as prompts, responses, and tool calls move through your application.
01
Inspect
Prompt, output, or tool payload
02
Score
Scanner-specific severity
03
Enforce
Allow, flag, or block
04
Resolve
Audit trail and remediation
Pricing plans
Start small, scale into governed AI operations.
Starter
For builders validating guardrails on early AI workflows.
$0for evaluation
- 1 organization workspace
- 2 active projects
- Input, output, and tool-call checks
- Basic logs and policy editor
Team
For teams preparing production AI agents and compliance workflows.
$49per project / month
- Unlimited policy iterations
- Red-team module and reports
- Resolution tracking and exceptions
- AI Policy Advisor with OpenAI analysis
Enterprise
For organizations that need governance, rollout support, and review controls.
Customannual plans
- Custom scanner and policy workflows
- Security review support
- Advanced reporting requirements
- Deployment and onboarding assistance
Interactive reviews
Ready to make your AI agent safer this week?
Book a practical guardrail review. We will map your riskiest AI workflow, show the exact scanner coverage, and outline the fastest path to production-ready remediation and reports.
Book Demo- Map your AI agent risk surface
- Review guardrail scanner coverage
- Walk through logs, reports, and remediation
- Plan a secure pilot rollout
Common questions
Common questions before adding AI guardrails.
Where does AgentGuard sit in an AI workflow?
AgentGuard scans inputs, outputs, and tool-call payloads before risky content reaches models, users, databases, or third-party tools.
Does AgentGuard store prompts?
The product stores hashes, scanner reasons, severity, and decisions. It is designed to avoid caching private prompt content in persistent logs.
Can teams manage remediation after a finding?
Yes. The MVP includes policy editing, red-team checks, reports, finding remediation, exceptions, retesting, and AI policy recommendations.