Skip to main content
xpander is the governed runtime for enterprise AI agents. It is built for the agents teams already use, Claude Code, Codex, Cursor and OpenCode, with any model, frontier or open-weight. Claude Code, Codex and OpenCode run inside xpander today; Cursor is not selectable yet. Agents run in the customer’s VPC, in xpander’s cloud, or air-gapped. With OpenCode and an open-weight model, nothing leaves the perimeter. xpander never writes the agent loop or the model. Teams assemble agents on top of them from approved skills and policies.

​
The agent runs on xpander, not beside it

Control is enforced as the agent acts, not reported after. Every call the agent makes, to a model or to a company system, passes through xpander:
  • Validated. The call is checked against the skills and policies the agent was given.
  • Credentialed at the moment of use. The credential is issued for that call. The agent never holds a secret.
  • Budgeted. The call counts against the organization’s and the agent’s limits.
  • Tied to a person. The call is recorded under the person who asked for it. Where systems allow, the agent acts with that person’s own permissions.
An action marked for approval waits for a named person, four hours by default; every skill call and model call lands in the task record under the person who asked.

​
What you get

Teams keep their tools. Nothing changes on a laptop. When someone asks Claude Code, Codex or OpenCode to run a job in the background, it runs on xpander in an isolated session with the same skills, approvals and record as any other agent, and answers in Xpander Chat, Slack, Teams, email, an API or on a schedule. See Start here. Security keeps control. Admins approve skills and models, set budgets and approval rules once under Settings, and every new agent starts inside them; Settings > Audit logs holds who changed what. Omni. The built-in agent: it finds agents and skills, runs and explains tasks, and assembles a new agent from approved skills when asked. One agent among yours, described on Xpander Chat.

​
What xpander is not

  • Not a harness or a model. xpander builds everything around the engine, never the engine.
  • Not a gateway. A gateway sits beside the agent; the agent runs on xpander.
  • Not a dashboard. Control is enforced at the moment of action, not after it.

​
Next

User guide

Launch a cloud session, use your desktop harness with the company’s skills, and hand jobs to the background.

Admin guide

Govern every agent. Deploy in your VPC, in xpander’s cloud, or air-gapped.