One control plane.
Total AI agent governance.
See Every Agent. Govern Every Action.
AgentPatrol™ is the In-Flow Control Plane for Enterprise AI.
Most governance tools watch an agent from the outside: a gateway that only sees calls to the LLM or your data. With AgentPatrol™, you can discover every agent already running in your enterprise, registered or not, and govern it in-flow, inside its own execution path, not from outside it.
You can govern down to the exact parameters of an action, not just block a tool outright, and do it across any framework your teams have already chosen, from LangGraph to OpenAI Agents to CrewAI, with more rolling out.
And you can prove, to the letter, what every agent did and why, with zero added cost or latency on the agent itself.
How AgentPatrol™ works
Register. Govern. Prove.
The same three moves, applied to every agent in your enterprise: the ones you registered and the ones you didn’t.
The Agent Registry and Tool Registry track every agent, every tool, and every credential automatically.
Action-level policy: allow, require approval, rate-limit, block. Enforced before the action executes, not after.
Immutable, replayable audit of every prompt, tool call, data touch and outcome, evidence an auditor will accept.
AI Agent Governance
Nothing runs ungoverned.
AI agents hold credentials, call tools, move money and touch customer data, and most are invisible to every control you already own. AgentPatrol™ is the control plane built specifically to see, constrain and prove what they do.
- Discover. Every agent, MCP server, tool and credential already running, including the ones nobody registered.
- Constrain. Action-level policy: allow, require approval, rate-limit, block. Enforced before the action executes.
- Prove. Immutable audit of every prompt, tool call, data touch and outcome, replayable for an auditor.
- Report. Map agent behavior to the EU AI Act, NIST AI RMF, and your internal risk framework.
Built For
Governance, risk, and AI leadership.
AgentPatrol™ stands on its own. No existing meshIQ or middleware relationship required. If you have AI agents running that you can’t see or control, that’s the exact problem it solves. If you have AI agents running that you can’t see or control, that’s the exact problem it solves.
CISO
Close the visibility gap.
Know every agent running in your enterprise, not just the ones your team registered. Shadow agents get discovered, not ignored.
Chief Risk Officer
Govern before it’s an incident.
Action-level policy enforced before execution, not a post-mortem after an agent has already moved money or touched data it shouldn’t have.
Head of AI / AI Center of Excellence
Say yes to AI, safely.
Give teams the freedom to build on any framework they choose, with one governance layer underneath, so speed and control aren’t a trade-off.
We have spent twenty-five years making the invisible traffic between enterprise systems observable, governable and accountable. AI agents are just the newest kind of traffic.
Greg DeaKyne, VP Product at meshIQ
Questions
Common questions about AgentPatrol™
AgentPatrol™ is meshIQ’s governance product for AI agents. It runs inside each agent’s execution path and checks every action against policy before it runs, deciding whether to allow it, ask the user to confirm, escalate it to a named approver, or deny it. Every decision is recorded so you can show exactly what an agent did and who approved it.
No. We don’t build or sell an agentic framework. Companies choose their own: LangChain, CrewAI, AutoGen, AWS AgentCore, or others, based on their own preference. AgentPatrol™ provides in-flow governance control across all of the popular frameworks available today, plus the same governance control plane across every agent, regardless of which framework built it.
No. AgentPatrol™ doesn’t require an existing meshIQ or middleware relationship. If you have AI agents running that you can’t see or control, that’s the whole problem AgentPatrol™ solves, on its own.
It’s discovered, inventoried, and held to the same default policy as every other agent until someone classifies it. Not silently allowed and not silently blocked. You decide what happens next, with full visibility into what it’s already done.
The goal is the opposite. Teams get the freedom to build on whatever framework fits the job; AgentPatrol™ handles the discovery, policy and audit trail underneath so speed and control stop being a trade-off.
Before. AgentPatrol™ evaluates an agent’s action against policy in-flow, resolving it to allow, hold for approval, throttle, or block, before it executes. That’s different from tools that only log what already happened.
A gateway governs what an agent sends. AgentPatrol™ governs what an agent does. Many agent tools run inside the agent or connect directly to databases, message queues and file systems, so those actions never pass through a gateway. Because AgentPatrol™ runs inside the agent, it can check every action however it’s carried out, pause the agent for human approval and resume it, and stop a runaway agent outright.
Policies look at the actual values in each tool call, not just which tool is being used. For example, a refund agent can process a $50 refund on its own, ask the user to confirm a $1,500 refund, and wait for a supervisor before issuing a $5,000 refund. Compliance teams set these rules centrally in the AgentPatrol™ console, so developers don’t have to code approval steps into each agent.
AgentPatrol™ currently supports LangGraph, the OpenAI Agents SDK, CrewAI, AWS Strands and Google’s Agent Development Kit, with more frameworks being added. Teams don’t need to standardize on a single framework, and one set of policies applies across all of them.
AgentPatrol™ masks personal and sensitive data before it reaches a language model and before it appears in the agent’s response. The real values still reach your system of record, so the agent can complete its task without the model ever seeing the data. Standard patterns such as card and Social Security numbers are predefined, and you can configure your own.
AgentPatrol™ doesn’t make any LLM calls of its own, so it adds no model cost. It tracks the token cost of each agent in real time, using the model provider’s own pricing data, and you can set token and iteration limits that stop an agent before it runs up an unexpected bill.
Factsheet
AgentPatrol™
AI agents are no longer just generating responses—they’re executing decisions. AgentPatrol™ governs those decisions while they happen.