🚀 RealTimeX v1.0.6 is live! Update now →
Using AI Agents
Setup

Setting Up AI Agents

The current RealTimeX agent setup flow is spread across a few connected surfaces. Start with the workspace Agent tab, then add runtime or plugin prerequisites only when your workflow needs them.

1. Open the workspace Agent settings

Go to your workspace, then open:

  • Workspace Settings > Agent

This is the main capability configuration surface for interactive @agent sessions in that workspace.

2. Configure Skills

In the Skills tab, choose which built-in plugin skills and imported skills the workspace agent can use.

This includes:

  • plugin-contributed skills
  • imported custom skills

If a skill is enabled here, the workspace agent can use it. If it is disabled here, the agent cannot rely on it even if the skill exists elsewhere in the instance.

3. Configure Agent Flows

In the Agent Flows tab, enable the flows you want available to the workspace agent.

Use this when the agent should be able to trigger reusable automation instead of handling everything as a pure chat task.

4. Configure MCP Servers

In the MCP Servers tab, expose the remote or local MCP servers the workspace agent is allowed to use.

For the full connection and management flow, see MCP Servers.

This is the main integration layer when you want agents to work with external tool ecosystems beyond built-in skills.

5. Check runtime prerequisites

Some capabilities depend on runtime setup outside the workspace Agent tab.

Review Agent Runtime when you need:

6. Configure behavior when needed

Agent capability is not only about tools. Some important behavior layers live elsewhere:

  • Workspace Settings > Chat for the active workspace System Prompt
  • Workspace Settings > Personality for file-based workspace identity and instruction files
  • Slash Commands for reusable manually-invoked prompt presets

For the reusable preset libraries themselves, see System Prompts and Personality.

7. Add browser capability when needed

If your workflow needs authenticated browsing or automation:

  1. Keep the Agent Browser built-in enabled.
  2. Configure browser sessions in Browser Tool.
  3. Test the flow from chat with @agent.

8. Add reusable authentication when needed

Authentication can live in more than one layer.

  • Use Credentials when workflows need reusable outbound secrets.
  • Use Agent Authentication when launched runtimes such as Claude Code, Codex CLI, Qwen, or Ambient Agent need provider API keys injected by RealTimeX.

For Antigravity CLI and Cursor Agent, use the runtime's own login flow first, then verify readiness from Agent Runtime instead of expecting an Agent Authentication profile.

This is separate from plugin-specific API key forms and separate from CLI login state in Agent Runtime.

9. Test with a small agent prompt

After configuration, test with a short task in the target workspace.

Examples:

  • @agent summarize the important documents in this workspace
  • @agent use the enabled browser workflow to inspect this page
  • @agent use the configured MCP tools to check the current system state

Setup checklist

  • Skills tab reviewed
  • Agent Flows tab reviewed if needed
  • MCP Servers tab reviewed if needed
  • behavior layer reviewed if needed
  • runtime prerequisites configured
  • browser sessions configured if needed
  • credentials added if needed
  • agent authentication added if needed
  • one real prompt tested successfully

Related guides

OverviewUsage