Discover agent boundaries
Identify the agent, its child processes, and the resources they can reach.
AI AGENT RUNTIME SECURITY
Know what agents do
on your systems.
Enforce their limits.
Discover what your agents can reach. See the processes, files, and connections behind each task. Define policies that allow, block, or require a person’s approval before an action goes ahead.
Discover · Visualize · Define · Enforce
The task“Update the payment service tests”
The agentCodex → bash → python
✓
Process startedpython test.py
File opened~/.aws/credentials
Connection attempted203.0.113.8:443
One task. The activity behind it.
Linked evidenceYour agents. Your frameworks.
-
Claude Code
-
Codex
-
Gemini CLI
-
LangGraph
-
CrewAI
KNOW THE ACTIVITY. SET THE RULES.
Understand the activity.
Define what’s allowed.
Follow an agent’s task through its processes, files, and network connections. See where its execution boundary ends and apply your policies to the actions inside it.
OmulO
Example view
Run activity
Illustrative evidence
PROCESSOBSERVATION
codexAgent root└─ bashChild process └─ python test.pyExecuted
The child process belongs to this run.
Process ancestry connects python to the agent. The task context follows the process tree.
Visualize each action
Trace a task through tool calls, processes, file access, and network activity.
Set the boundaries
Define what an agent can do. Allow routine work, block prohibited actions, and require human approval when your policy calls for it.
HOW IT WORKS
From activity
to policy.
A tool call is only the start. OmulO follows the system activity behind it and applies the boundaries you define for the agent.
Talk through your setup-
01
Discover the execution boundary
Identify the agent and task, then follow child processes and the files, tools, and network destinations within their reach.
-
02
See the action chain
Connect task context with independent system observations. Visualize the sequence of actions and the process responsible for each one.
-
03
Define the policy
Set rules for file access, commands, and network destinations. Decide which actions are allowed, denied, or sent to a person for approval.
-
04
Enforce the decision
Apply the matched rule before the action proceeds. Hold approval-required actions for a reviewer and keep the decision with the action record.
POLICY DEFINITION & ENFORCEMENT
Let the agent work.
Keep people in control.
Set the boundaries once. OmulO checks agent actions against your policies and brings a person into the decision when approval is required.
Define what’s allowed
Scope rules to the agent and the resources it uses. Set permissions for commands, file access, and network destinations.
Require human approval
Hold selected actions before execution. Give the reviewer the task, process, affected resources, and matched rule to approve or reject.
Keep the decision with the action
See the matched policy, approval decision, and enforcement result alongside the activity that triggered them.
payment-service agent
Boundary: payment-service workspace
| Action | Policy |
|---|---|
| Edit project filesInside the workspace | Allow |
| Read credential filesOutside the agent’s boundary | Deny |
| Deploy to productionChanges a protected environment | Require approval |
HUMAN IN THE LOOP
Production deploy held for review
The reviewer sees the requested action and its context before deciding whether it can proceed.
Matched ruleProduction changes need approvalAction stateWaiting for a person’s decision
EVIDENCE YOU CAN CHECK
Enforcement starts
with reliable evidence.
To enforce a rule, you need to know what happened—and where visibility ends.
Observed means observed.
What an agent declares and what the kernel records are separate evidence sources. An inference should never look like a fact.
Missing coverage stays visible.
Capture loss and policy exclusions are accounted for. Missing telemetry is never treated as proof that nothing happened.
Collection is explicit.
Prompt content, command arguments, and environment data need clear capture settings. Environment collection is disabled in the shipped configuration.
PRIVATE EARLY ACCESS
Set the boundaries
for your agents.
Let’s review what your agents do on your systems and where you want policy boundaries.
Bring a workflow, the resources it touches, and the actions that need a person’s approval. We’ll walk through the boundaries and policies with your team.
hello@omulo.aiA FEW DETAILS
Before we talk.
How does OmulO discover an agent’s boundary?
OmulO connects the agent and its task to the processes it starts, the tools it invokes, and the files and network destinations those processes access. That execution map gives your team the context to define and enforce the agent’s limits.
What can I put in a policy?
Define rules for commands, file access, and network destinations within an agent’s scope. Choose whether an action is allowed, denied, or requires human approval.
How do human approvals work?
When a rule requires approval, OmulO holds the action for a reviewer. The reviewer sees the task, requested action, affected resources, and matched rule, then approves or rejects it. The decision stays linked to the action record.
Where does OmulO run?
OmulO runs on Linux hosts, including agent servers, CI runners, virtual machines, and container hosts. We’ll review your agent workflow and deployment with your team.
How do we control sensitive content?
Use explicit capture settings for prompts, command arguments, paths, and tool inputs. Environment collection is disabled in the shipped configuration. Review the capture policy with your team before connecting a workflow.