Enterprise AI agents belong in the cloud | xpander

What changes when your local AI agents move to the cloud

Desktop was a great place to start. Cloud-deployed AI agents are governed, shared, and always on.

Local agents on your desktop

Cloud-deployed agents with xpander

Execution permissions

• Ask permission only when the model feels like it
• Follow deterministic approval gates, per skill

Model neutrality

• Run on one vendor’s model
• Run on any model behind one shared gateway, no rewrite required

Credentials

• Keep secrets in local config files
• Pull credentials from a shared vault at runtime, invisible to the model itself

Availability

• Stop working when the laptop closes
• Run on a schedule, a trigger, or on demand

Auditing

• Access whatever they can reach, unchecked
• Log every action, cost, and approval to a full audit trail

Internal systems

• Can’t connect to systems behind the firewall
• Connect to the systems already behind your firewall

Team access

• Tied to one person’s laptop
• Run for the whole team once published, scoped per user

See xpander in action

How local AI agents become governed agents

Powered by Omni, xpander’s built-in AI engineer

01

Connect Omni where you build

Connect Omni by xpander to wherever you build, Claude Desktop, Claude Code, Codex, or any MCP-compatible client. Omni authenticates as you and reads your local setup.

02

Ask it to deploy what you built

Omni reads the skills, instructions, and agent setup you wrote locally and redeploys them on xpander, including skills, tools, connections, and the agent itself, ready to run in the cloud.

03

It comes up governed

Named identity, per-skill approval gates, credentials from the vault, budgets, and a full audit trail, live from the first run and reachable from every channel your team already uses.

DEPLOYED

04

Collaborate from anywhere

Work with your agent from wherever you already work: Claude, Slack, Teams, or email. People and agents share one thread.

Start from the tool you’re already using

Claude DesktopClaude CodeCodexCursorOpenClaw / Hermes

1. Add Omni from the Claude connector directory: claude.ai/directory/omni-by-xpander
2. Authenticate - Omni acts as you, scoped to your permissions.
3. Send the first message: "Omni, look at the skills and agents in this project and deploy them to xpander."

Skills written as SKILL.md files can also be uploaded directly - no connector required.

What actually moves

On your desktop, the agent is one opaque setup only its builder can see. Deployed, it unpacks into first-class constructs - each one visible, permissioned, and audited.

ON THE LAPTOP

local agent

ON XPANDER

MCP servers

LOCAL + REMOTE

• Skills
• APPROVAL GATE
• Organizational connectors
• PER-USER
• System prompts
• VERSIONED
• Tests
• MOCK DATA
• Human in the loop
• ENFORCED
• Approval flow
• IN-THREAD
• Model selection
• ANY MODEL

Nothing about your workflow changes

The laptop stops being production. You don’t. Keep developing in Claude Desktop or wherever you build - and keep talking to your agent from there too. Deployed agents stay reachable from Claude, ChatGPT, Slack, Teams, email, and the web. The next agent you build takes the same deploy step, and the one after that. The desktop is the dev environment. xpander is prod.

Frequently asked questions

Can the agents I built in Claude Desktop run in production?

Yes. Connect Omni where you build, ask it to deploy what you have, and the agent comes up on xpander - governed, shared, and audited - while you keep developing on the desktop.

What happens to the skills I wrote locally?

Omni reads them and converts them into xpander skill constructs. Each one can carry its own approval gate, credentials, and budget. SKILL.md files can also be uploaded directly.

Do I have to stop using Claude or Codex?

No. The desktop stays your development environment, and deployed agents remain reachable from Claude, ChatGPT, Slack, Teams, and the web. The agent moves; you don't.

Is this a migration?

It's a deploy step. Migration implies leaving a tool behind. Here the laptop stays where you develop and test; xpander is where the agent runs for the company. If you searched for how to migrate Claude agents to production, this is the page - the workflow just doesn't ask you to give anything up.

Which desktop tools does this work from?

Any MCP-compatible client: Claude Desktop, Claude Code, Codex, Cursor, OpenClaw, Hermes, ChatGPT, and whatever you adopt next.

What does "governed" actually mean here?

Deterministic human-in-the-loop gates per skill, credentials injected from a vault at runtime, named agent identity with per-user permissions, budgets per task and per agent, and a full audit trail.