Capabilities · Agent Guardrails
Give agents the map
before they act.
See what an AI agent can reach and what an action would affect, from your live graph, with a person approving anything that changes your cloud.
What's Included
2 jobs and 5 quick assessments for agent guardrails.
Jobs do the work. Quick assessments grade it on a fixed checklist, with the evidence behind every check.
- 01
We get you started.
Every job and quick assessment installs free with Operator. Copy a prompt and run it yourself, or have us or a partner run it with you.
- 02
Better on the graph.
Answers come from live state, owners, reach and cost in your Cloud Intelligence Graph™, not a guess from code.
- 03
Better as a team.
On Pro, everyone works from one shared graph. Findings, plans and context pass between people, and a correction made once counts for everyone.
Jobs 2
- Accounts and identitiesEvery account matched to an approved, read-only Identity.
- Security hardeningA ranked plan, each fix written against the code that declares it.
Quick Assessments 5
- Security postureEstate-wide controls, with the four checks below rolled up.
- Least-privilege accessWho or what can act on a Service, and roles that can escalate.
- ExposureWhat is reachable from the internet, against the front doors you expect.
- SecretsStorage, rotation and readers, from names and dates only, never values.
- Pipeline supply chainWhether a Pipeline or pull request could be used to reach Production.
The Problem
Nobody approves autonomy they can't see.
Agents already act with their operator's permissions. The people who answer for production want answers first.
- What can the agent reach? Credentials often reach far more than the task needs.
- What would this action affect? There is rarely a way to see the consequence before it runs.
- Who answers for it? Ownership and approval are unclear until after the fact.
What the Graph Checks
The facts an agent's actions are checked against.
OpsCanvas doesn't contain or sandbox agents. It gives them, and the people approving them, the confirmed map.
- What an agent can reachAvailable The resources and environments its access touches, and who owns them.
- Blast radius before an actionAvailable What a proposed action would affect, before it runs.
- A person in the loopAvailable Every MCP tool is marked read or write, so anything that changes your cloud waits for approval.
- Agent action controlsComing 2027 Actions checked against the graph before they run, with a record of each one. Part of Enterprise.
Try It Yourself
Copy the prompt. Run it in your AI tool.
Run it in your AI tool with OpsCanvas installed. These are real prompts from the OpsCanvas Help examples.
Sample Jobs
- Managing accounts and credentials Runs: opscanvas-starter-accounts
Which of our accounts can I reach through OpsCanvas, and which am I missing?- With the graph
- Every account the estate names, matched to an approved identity or flagged. A missing one is approved only after a live login succeeds, and no credential is read or stored.
In the Session
Your AI acts only through identities you approved.
A model only ever sees the identities listed in OpsCanvas. Nothing it finds on the workstation is usable until it is approved.

Better With the Graph
AI agents are only as safe as their guardrails and accurate as their context.
| An agent alone | With the Cloud Intelligence Graph™ | |
|---|---|---|
| What it can reach | Whatever its credentials allow | Known, with owners attached |
| What an action affects | Found out afterward | Shown before it runs |
| Who approves | Unclear | A person, with the evidence in front of them |
On Pro, the whole team checks agent actions against one shared graph. Enterprise adds agent action controls across the estate, coming 2027 with design partners now. Talk to us about design partnership →
Agent Guardrails
Give your agents the map today.
Free with Operator, in the AI tool you already use.