Agent runtime

Start persistent agent work from ExecWarden.

ExecWarden can run the agent loop without giving it a general-purpose machine. Add a customer-configured VM, Docker environment, or local machine only when the task needs files, shell, git, browser, tests, or private-network access.

Identity and tool activity

Keep the caller, its grants, and tool-call results inspectable.

Each named Agent identity keeps its grants and tool-call records together. The activity log does not claim visibility into commands or credentials used directly inside a connected execution environment.

Maintenance Agent identity with a scoped GitHub grant
The named caller carries its own grant scope, mode, and lifecycle.
Recent GitHub activity through ExecWarden showing successful, failed, and approval-paused calls
The identity view keeps several outcomes visible without presenting local machine work as gateway activity.

What you can do

Keep the run visible without assuming it has a machine.

Start with limited JavaScript, public HTTP where enabled, and granted tools. Add a customer-configured VM, Docker environment, or local machine only for tasks that need shell, files, git, browser, tests, or private-network access.

Persistent sessions

Keep task state attached to the run.

Hosted run state keeps the transcript, tool calls, approvals, and status together whether or not the task uses connected compute.

Tool access

Give the session only the tools it needs.

Attach GitHub, package-backed APIs, and imported MCP tools through the Agent identity. Configure execution capacity separately when the task needs a machine.

Compute when needed

Connect a machine for machine-shaped work.

A customer-configured VM, Docker environment, or local machine supplies the shell, files, git, browser, processes, and network access configured for the task.

Examples

Hosted tasks with and without connected compute.

1Research public information and call granted APIs without attaching a machine.
2Attach customer compute for a repo investigation, tests, and branch output.
3Move a trusted task toward a workflow after the manual run is useful.

Get started during beta

Start with one concrete task and a clear access plan.

Use the app during beta, or open the docs and choose the first agent task you want to control.

Start an agentRead the security model