Definition
What is BobSentry?
BobSentry is runtime authorization infrastructure for consequential AI-agent actions. It evaluates whether a specific action should proceed right now—ALLOW, REQUIRE_APPROVAL, or BLOCK—and creates signed decision evidence. BobSentry does not execute the downstream action.
Runtime authorization, not orchestration
BobSentry evaluates normalized action metadata and returns ALLOW, REQUIRE_APPROVAL, or BLOCK before execution. Sensitive payloads can remain in your environment.
Learn more on Product, try the Developers sandbox path, or review the Security model. Coding-agent teams should see Dev Guard.
Frequently asked questions
What problem does BobSentry solve?
AI agents can propose consequential actions against real systems. Permissions answer what an agent can access; they do not decide whether a specific action should proceed now. BobSentry provides runtime authorization—ALLOW, REQUIRE_APPROVAL, or BLOCK—with signed decision evidence before the integration executes.
How does it work?
Your agent or gateway sends normalized action metadata to the Runtime Evaluate API before calling the external system, and receives a decision. In COOPERATIVE mode, the integration must honor that decision. Approvals, when required, are bound to the exact proposed action.
What decisions does it return?
ALLOW (policy permits the requested action), REQUIRE_APPROVAL (human authorization is required before the integration should proceed), or BLOCK (policy denies the requested action).
What data does BobSentry receive?
Normalized action metadata. Sensitive payloads, source code, PHI, credentials, and customer data can remain in the customer environment.
How is it different from permissions?
Permissions describe what an agent can access. BobSentry authorizes whether this specific consequential action should proceed now, given policy, target, and context.
How is it different from orchestration?
Orchestration frameworks schedule and chain agent steps. BobSentry is an independent authorization boundary immediately before an action reaches a real system—not a workflow engine.
How is it different from observability?
Observability records what happened. BobSentry decides whether an action may proceed, and produces signed decision evidence of that authorization.
How do approvals work?
Humans should not approve every AI action—they should approve the actions that matter. When the decision is REQUIRE_APPROVAL, an operator authorizes or rejects the exact proposed action. Approval means authorization to proceed; BobSentry does not execute the downstream action.
What actions can be protected?
Any consequential agent action you can evaluate before execution—external messaging, financial operations, EHR writes, file and Git changes, dependency installs, shell commands, infrastructure applies, and deployments.
Who is it for?
Teams operating AI agents that can take consequential actions—AI infrastructure founders, CTOs, platform engineers, and technical security owners—especially design partners integrating a real pre-execution call site.
How can developers test it?
Create a sandbox account at the BobSentry application, register an Agent, issue an Agent-bound runtime credential, call the Runtime Evaluate API, and inspect signed decision evidence. See the Developers page for the path.
Runtime API: https://runtime.bobsentry.com