AgentGuard
Put guardrails on your agents inputs and outputs
AgentGuard reviews an agent's prompt and tool list, rates the risk of each tool call, and suggests input/output filters plus an EU AI Act mapping — so an agent cannot silently exfiltrate data or take an unguarded action. It starts at $29/month.
AgentGuard reviews an agent's prompt and tool list, rates the risk of each tool call, and suggests input/output filters plus an EU AI Act mapping - so an agent cannot silently exfiltrate data or take an unguarded action.
What it covers
- Tool-call risk rating
- Input/output filters
- EU AI Act mapping
- Blocklist suggestions
Coverage at a glance
- Starts at $29/month
- Rates risk per tool call
- Maps risks to EU AI Act obligations
How does it rate tool-call risk?
It reads the agent's tool list and rates each call by what it can touch — send mail, run code, read files, call APIs — so the dangerous edges are named, not assumed safe.
What filters does it suggest?
For each risky edge it proposes input and output filters and a blocklist, turning a vague 'be careful' into concrete checks you can implement.
How does the EU AI Act mapping help?
Each flagged risk ties to a relevant AI Act obligation, so your guardrail work is documented for governance instead of living only in a developer's head.
Is AgentGuard a runtime firewall?
It is a review-and-suggest layer you run before deployment. Pair it with a runtime guard so the suggested filters are actually enforced when the agent runs.
Where AgentGuard stops short of compliance
AgentGuard does one thing well: Put guardrails on your agents inputs and outputs. It watches or grades agents, but a dashboard is not a kill switch. Keep a human accountable for what the agent actually does.
Questions about scope and limits
What does AgentGuard do?
AgentGuard reviews an agent's prompt and tool list, rates the risk of each tool call, and suggests input/output filters plus an EU AI Act mapping so an agent cannot silently exfiltrate data or act unguarded.
How much does AgentGuard cost?
Plans start at $29/month (billed $290/year). See the live tool for the current plan before reviewing a production agent.
Who is AgentGuard for?
AgentGuard is built for teams shipping or operating AI agents who need this exact job done: Put guardrails on your agents inputs and outputs.
How does it rate tool-call risk?
It classifies each tool by what it can do — mail, code, files, external APIs — and flags the calls most likely to cause data loss or an unguarded action.
Does it map to the EU AI Act?
Yes. Each risk ties to a relevant AI Act obligation, so the guardrail work is governance-ready rather than informal.
Is it enough on its own at runtime?
It reviews and suggests; you still enforce the filters with a runtime guard. Use AgentGuard to decide what to block, then a runtime layer to block it.
Is AgentGuard enough to be compliant on its own?
Not a substitute for a professional security audit or a formal compliance assessment. Use it to catch obvious gaps before production; regulated systems still need expert review.
Built by Xingliang Li · LX Micro-SaaS Factory — a one-person, curated set of tight-vertical AI tools. How tools are selected →