An agent sees what the person who runs it can see. Nothing more.
Row- and column-level Cedar policies, applied in real time on every query. Agents inherit the permissions of the user who starts them. Every access is logged with the reason it was allowed or denied.
Six answers for the security review.
Row and column policies
Access is written as Cedar policies at row and column level and applied in real time, on every query, for every caller.
Agents inherit the person
An agent runs with the permissions of the user who starts it. It cannot see a row that person could not see, whatever the prompt says.
Every access, with its reason
Each read is logged with the policy that allowed or denied it. A reviewer reads why, not just what.
Isolated per tenant
Customer data is physically separated per tenant. Only schemas and settings are gated in the application.
Models under policy
You choose the model providers. A policy decides who is allowed to change them.
Where it runs
Your cloud, ours, or on-prem, including air-gapped with scheduled updates. SSO through OpenID Connect.
How a request is decided.
Who is asking
The person signs in through your OpenID Connect provider. An agent carries the identity of the user who started it.
Which rows and columns
Cedar policies decide which rows that identity can see and which columns it can read, at query time, below the model.
The query runs
Only within what the policy allows. No instruction in a prompt can widen it, because the model never holds the permission.
The decision is logged
Allow or deny, with the policy that made the call, for people and agents in one log.
The last line is a person, not an agent: changing the model provider is itself a policy decision.
Physically isolated per tenant, or never leaving your network.
In Ward’s cloud, customer data is physically separated per tenant; only schemas and settings are gated in the application. Or run Ward in your own cloud, or on-prem, including air-gapped with scheduled updates. Live databases are queried in place through the proxy on a read-only replica, so they are not copied at all.
For compliance status, sub-processors and the documents a reviewer will ask for, see the trust center and the evaluation packet.
What a security reviewer asks first.
With Cedar policies at row and column level, applied in real time on every query for every caller. A policy decides which rows a person can see and which columns they can read, and the same rule applies whether they ask in the console or through an agent.
Exactly what the person who started it can see. Agents run with that user's permissions, and policy is applied below the model, so no instruction in a prompt can widen it.
Yes. Each access is logged with the policy that allowed or denied it, so a reviewer can read why a request was allowed, not only that it was.
Yes. Customer data is physically isolated per tenant. Only schemas and settings are separated at the application layer.
You choose the model providers, and a policy controls who is allowed to change them. A user without that permission cannot switch models.
SSO through OpenID Connect providers. Keycloak is in use today, and adding another OpenID Connect provider is a short configuration task.
For your reviewers.
The architecture
Where the model sits, where the math runs, and where policy applies.
Read it →Compliance and documents
Status, sub-processors, and the documents your reviewer will ask for.
Open it →The agent audit log
What a complete agent audit record contains, field by field.
Read the schema →The proof chain
Every figure linked to its source rows and checked before it returns.
Try it →Your security review starts with a short list of questions.
Architecture, policy model, access log, isolation. On the table before you sign.
Read-only to start · your LLM keys · SOC 2 Type II underway · or book a call directly
Find out what your data has been hiding.
Tell us about your operation. We’ll show you the problems Ward catches, and the ones your current tools miss.