A security layer around every agent

Protoguard is Metorial’s security layer for AI agents. It reviews incoming messages and tool requests before an agent acts, catches prompt injection, watches for provider schema changes, and blocks anything outside your policies before it reaches your systems.

Get startedBook a demo

Set up Protoguard in three steps

Configure the monitors you need, review the alerts they raise, and enable tool filters that block unsafe requests before they reach connected systems.

Screenshot of the Protoguard monitors configuration panel.

  1. Turn on the monitors that fit your agents, from prompt injection to provider schema changes, and set what counts as risky.

Defense in depth

Runs on isolated infrastructure

Protoguard's reviews sit on top of Metorial's runtime isolation, so even a request that slips past a filter stays contained within its own boundary.

Enclaves

Providers and custom code run in isolated enclaves. Every action is sandboxed and contained.

  • Dedicated runtime boundaries
  • Custom networking per provider
  • Safe for first-party, untrusted, and custom providers

Firewalls

Firewalls control which destinations a provider can reach. Outbound and inbound traffic always stays within the rules you set.

  • Outbound traffic controls
  • Allowlist trusted destinations
  • Per-provider network policy

Vault and KMS

Secrets live in Metorial Vault, encrypted with KMS and resolved at the gateway, never shared with the model, client, or user.

  • KMS-backed secret storage
  • Credentials resolved at runtime
  • Nothing sensitive reaches the model

Visibility in production

Give your security team a clear view

When a monitor fires, your team needs to know what happened and act fast. Protoguard turns flagged activity into alerts with context, surfaced in one place and recorded for review.

Alerts with context

Every alert shows what triggered it, which agent was involved, and the user identity behind the request.

  • The agent and user behind each event
  • What the monitor flagged and why
  • No digging through raw logs

One place to review

Message reviews, policy checks, and blocked requests all land in the Protoguard console, so your team works from a single view.

  • Flagged activity in one place
  • Filter by agent, user, or provider
  • Blocks and reviews side by side

A recorded audit trail

Reviews, alerts, and blocked requests are recorded, giving security and compliance a clear history to audit.

  • Every decision recorded
  • Searchable activity history
  • Evidence for compliance reviews

What Protoguard watches

Every interaction runs through the same checks, so unsafe instructions, risky tool use, and provider changes are caught before an agent acts.

  1. Prompt injection

    Incoming messages are scanned for instructions that try to override your agent before it acts.

  2. Schema changes

    Protoguard tracks when a provider changes its tools, so updates that could break or hijack behavior are caught early.

  3. Policy violations

    Every tool request is matched against your filters and access policies, with the user identity behind the agent.

  4. Data exfiltration

    Outbound activity is watched so sensitive data cannot leave through a connected system.

  5. User identity

    Each request is tied to the user behind the agent, so policy checks run with the right context.

  6. Firewall filters

    Tool requests are matched against your allowed filters before they can reach a connected system.

  7. Credential scope

    Secrets stay out of model context and are resolved at the gateway when a request is allowed.

Put a security layer around your agents

See how Protoguard reviews, monitors, and blocks agent activity before it reaches your systems. Book a demo or get started today.

Book a demoGet started

Explore the Protoguard platform

Learn more about Protoguard and how it secures AI agents in production.

Explore Protoguard

See what Protoguard can do, in detail.

Prompt Injection Monitoring

Observability

Metorial Protoguard monitors every tool call, message, and execution for prompt injection attacks in real time. Attacks on your agents are detected before they do damage.

Schema Change Monitoring

Observability

Metorial version-controls every integration and skill and tracks changes to tools, configs, and schemas. An unexpected change is caught and alerted before it breaks an agent that depends on it.

Monitors and Alerts

Observability

Metorial monitors your integrations, traffic, and logs and alerts you when something needs attention. You can catch problems like schema changes and errors early.

Firewalls

Infrastructure

Metorial's virtual firewalls govern inbound and outbound network traffic for every integration and MCP server. You can enforce per-workload network policy, block exfiltration, and monitor all traffic.

Policies + Access Control

Governance

Metorial gives you granular, role-based access control over users, agents, and admins, with policies across skills, integrations, and agents, built on your existing identity.

Enclaves

Infrastructure

Enclaves are the fast, secure sandboxes that run every integration on Metorial. Integration code is always isolated and monitored, down to every user, connection, and tool call.

Metorial Vault

Infrastructure

Metorial Vault stores the credentials your integrations need securely and uses them without exposing them, with encryption keys managed by Metorial or by you and every operation audited.

KMS

Infrastructure

Metorial supports secret encryption backed by AWS KMS, including your own KMS keys. You control the encryption of secrets stored in Metorial, with full audit logs.

Frequently asked questions

Common questions about Protoguard.

  1. What is Protoguard?
    Protoguard is Metorial’s security layer for AI agents. It reviews messages and tool requests, watches for risky activity, and blocks anything outside your policies before it reaches connected systems.
  2. Yes. Protoguard reviews incoming messages for unsafe instructions through prompt injection monitoring before an agent ever acts on them.
  3. Protoguard uses schema change monitoring to catch when a provider changes its tools, so updates that could break or hijack agent behavior are flagged before they reach production.
  4. Each tool request is checked against your firewall filters and access policies. Requests that do not match are blocked before they reach the system, and every decision is recorded.
  5. Yes. Monitors and alerts raise a notification when agent activity looks unusual, so your security team sees risky behavior as it happens.