Named agent access
Connect coding agents such as Codex, Claude Code and Cursor through identifiable access.
Local access control for named AI agents. Define what an agent may use, for which task, under which limits.
Open beta programme · Apple-silicon Macs
Current availability and requirements are listed at vardra.dev.
An example grant / illustration
A local decision before execution.
An agent needs access to do useful work. The next question is which identity may use a credential, what it may do, how long access lasts and how you can stop the next use.
Give each connected coding agent its own identity, grants and attributable decisions.
Choose the secret field, action, expiry, lease duration and use limit. Scope the access to the task.
Use Auto-allow for permitted routine work or Ask-first when a local approval is required.
Read the local audit and revoke a grant or agent to deny its next broker use.
The vault, grants, approval inbox and access audit belong to a local workflow on your Mac.
Connect coding agents such as Codex, Claude Code and Cursor through identifiable access.
Define the capability and its limits rather than leaving a credential in standing agent context.
Keep the decision and record on your Mac, with a clear path to review and revoke.
Save the one-time Emergency Kit when creating the vault. Recovery depends on material you retain.
The trusted child process holds the credential while it runs. Vardra withholds arbitrary child output and returns exit metadata.
This is an access policy boundary, not a process sandbox or an exfiltration defense. Same-user processes remain within the operating-system account's trust boundary. The readable vault opens on your device.
Keep your Emergency Kit safe. Vardra cannot reset the master password for you; recovery comes from material you control.
Read the current product informationFind the current Mac download, setup requirements, beta terms and product information on Vardra's own website.
App support: [email protected]