Product reference · Agentic harness
Rules, permissions and scope
An assistant only acts with proper permission, and everything it does stays inside the venue’s own availability checks and limits.
~2 min read
When you'd use it
To understand the guardrails before letting the agent act.
How it behaves
Access to the graph and the service databases is read-only, and any write waits for your confirmation, so nothing changes behind your back. The agent is kept from revealing secrets or internal configuration, and everything it works with is your own data on your own account.
Guardrails apply on every step, so heavier or longer-running work stays within the same controls as a quick query.
Configuration and options
- Read-only access
- The agent reads the graph and databases; it does not write on its own.
- Sign-off on writes
- Any change to data waits for your confirmation.
- Guardrails
- Controls on every step, including background work.
Good to know
Read-only by default; a write only happens with your sign-off.
It is your own data on your own account throughout.