How To Use AI Agents

Use @agent when you want RealTimeX to work through a task over multiple steps instead of answering with a single chat completion.

Start an agent session

In a workspace chat, type:

@agent your request here

Example:

@agent review the files in this workspace and tell me what needs attention first

Continue the session

Once the agent session starts, keep chatting normally in the same thread.

You do not need to repeat @agent on every follow-up message while the session is active.

End the session

To end the agent loop early, use:

/exit

💡

The current product behavior uses /exit as the explicit end-session command.

What the agent can draw from

What your agent can do depends on workspace and runtime setup, not just one fixed internal tool list.

Depending on configuration, an agent may use:

workspace documents and search
workspace system prompts
workspace personality files
plugin skills
imported skills
agent flows
MCP servers
Agentic CLIs
working directories
browser sessions
reusable credentials
runtime-injected provider authentication

Good prompt patterns

Workspace analysis

@agent summarize the key themes across the documents in this workspace
@agent tell me which files I should read first before this meeting

Browser-backed research

@agent compare these product pages and extract the differences
@agent use the browser session to inspect the logged-in dashboard

Runtime-aware work

@agent update the wiki project in my working directory with a clean summary
@agent use the enabled MCP tools to verify the latest deployment state

Output generation

@agent turn this data into a chart and explain the trend
@agent prepare a file I can download from this result

How to tell a session is active

When the agent starts, RealTimeX shows a handoff message indicating the agent was invoked and the chat has switched into agent mode.

From that point on, you are in an agent session until:

the task completes
the workflow exits naturally
you use /exit

Common mistakes

Treating the agent like a one-shot tool picker

Current RealTimeX agents are broader than the old built-in tool catalog. If the agent seems limited, check configuration first instead of assuming the feature is missing.

Expecting browser or file access without setup

If the task needs browser sessions, working directories, reusable credentials, or runtime-injected provider auth, those prerequisites must already be configured.

Repeating `@agent` unnecessarily

Use @agent to start the session. After that, continue naturally unless you intentionally want to trigger a fresh invocation later.

If the agent cannot do what you asked

Check these layers:

Workspace Settings > Agent
Agent Runtime
Plugins
Credentials or Agent Authentication if auth is involved

Related guides

Setup Introduction