Runtime authorization boundary
BobSentry sits immediately before execution—after an agent proposes an action, before tools, APIs, or infrastructure are mutated. It returns ALLOW, REQUIRE_APPROVAL, or BLOCK.
Security
Designed as an independent runtime authorization boundary
BobSentry evaluates normalized action metadata with deterministic policy decisions. Metadata-only by default: raw business payloads can remain in your environment while BobSentry receives the normalized action context needed for authorization. Every recorded runtime decision produces signed authorization evidence.
Data-flow diagram
Customer environment
- AI agent
- Sensitive payload
- Local classification / redaction
Normalized action metadata
BobSentry
- Deterministic policy evaluation
- Human authorization
- Signed decision evidence
Deterministic policy evaluation
Policies evaluate action metadata, target, context, and risk to produce a deterministic decision. High-risk actions can require human authorization without stopping every autonomous workflow.
Metadata-first architecture
BobSentry is designed to evaluate normalized action metadata rather than requiring raw source code, PHI, financial records, credentials, or full prompt histories. Integrations can redact or summarize sensitive context before the evaluate call.
Sensitive payload handling
Customer-controlled agents and services retain raw business payloads by default. BobSentry receives the normalized action context required for authorization. Caller-supplied metadata is not guaranteed to be free of sensitive information—that depends on integration design.
Transport security
Runtime API traffic uses HTTPS. Agent-bound runtime credentials authenticate evaluate callers; credentials should be stored and rotated according to your secret-management practices.
Signed authorization evidence
Each decision can be exported as signed JSON (and PDF where enabled). Evidence captures the agent, requested action, policy evaluation, decision outcome, whether authorization was required, who authorized when applicable, and cryptographic integrity. Execution outcomes are recorded separately from authorization—audit-ready records of authorization decisions, not a compliance guarantee.
Action hashing
Action hashes bind approvals and evidence to the proposed action so an authorized decision cannot silently apply to a different payload.
Tenant isolation
Evaluation, policies, approvals, and evidence are scoped per tenant. Cross-tenant access is not part of the product model.
Credential handling
Agent-bound evaluate-only credentials authenticate Runtime API callers. Do not embed production credentials in public clients. Rotate or revoke compromised credentials immediately.
Failure-mode configuration
Integrations should define fail-closed or fail-open behavior for unreachable evaluate calls based on risk tolerance. Prefer fail-closed for consequential production actions. In COOPERATIVE mode, the integration must honor ALLOW, REQUIRE_APPROVAL, and BLOCK.
Certifications and compliance posture
BobSentry is designed to support regulated workflows. We do not claim completed SOC 2, HIPAA, GDPR, ISO 27001, or other certifications on this site unless evidence is published. Contact us for current status.
Responsible disclosure
Report security issues to hello@bobsentry.com with enough detail to reproduce. Do not include real customer data in reports. Contact hello@bobsentry.com.