Prompt and content risk screening
Identify prompt injection patterns, untrusted instructions, hidden directives, and risky content before an agent acts.
Agent Guard 1.0.1 — controlled release Produit distinct
Guard what AI agents can see, call, and send.
Agent Guard is a separate browser-agent security product. On sites a user explicitly authorizes, it evaluates supported actions against intent and local policy, can request approval or block supported commits, and preserves a minimized local audit trail.
Separate codebase Policy-first Auditable
Prompt & context
Agent Guard policy gateway Allow Deny / redact
AI agent
Tools / MCP
Outbound data
Human approval
Produit distinct
View the SiteEgress product
Agent Guard is not the SiteEgress Chrome extension.
SiteEgress inspects browser-visible website privacy behavior. Agent Guard supervises supported browser-visible agent actions on sites the user authorizes. The products have separate codebases, permission models, interfaces, privacy policies, and release paths.
Intended capabilities
Runtime controls designed for agentic systems
The following capabilities describe the controlled-release scope. Coverage is limited to supported browser-visible channels and is never presented as complete protection against every agent or prompt-injection risk.
Tool and MCP policy
Define which tools, MCP servers, methods, and resource scopes an agent may use for a given task or identity.
Data egress controls
Inspect and constrain what information may leave trusted boundaries through tools, APIs, messages, or generated files.
Human approval gates
Require explicit review for high-impact actions such as payments, destructive changes, external communications, or secret access.
Secrets and PII safeguards
Detect sensitive values and apply masking, denial, substitution, or approval policies before tool execution.
Auditable decision traces
Record policy inputs, decisions, approvals, tool calls, and outcomes in a reviewable security trail.
Product architecture
A policy gateway between intent and action
Agent Guard is designed as a control point around an agent—not as another general-purpose chatbot. Policies should be testable, decisions explainable, and failures visible.
Testable policies
Explainable decisions
Visible failures
Prompt & context
Agent Guard policy gateway Allow Deny / redact
AI agent
Tools / MCP
Outbound data
Human approval
Designed for
Who Agent Guard is for
Agent platform teams
Add a consistent policy layer across internal agents, tools, and MCP integrations.
Security and governance
Review high-risk actions, data flows, exceptions, and policy coverage.
SaaS builders
Offer customer-facing agents with clear scopes, approvals, and auditability.
Automation teams
Protect workflows that touch email, finance, support systems, code, or production infrastructure.
Diffusion contrôlée
Request controlled release
Controlled release access
Agent Guard 1.0.1 is distributed separately for controlled evaluation. Access does not imply production certification, complete channel coverage, or a guarantee against every agentic threat.
Help define practical guardrails for agentic systems.
Tell us which agents, tools, and data boundaries you need to govern.