AI Agents

AI Agents in RealTimeX are session-based assistants that can do more than answer a prompt once. When you invoke an agent, it can combine workspace context, enabled skills, flows, MCP servers, and runtime tools to complete multi-step work.

How agent sessions start

In a workspace chat, start an agent session with:

@agent your request here

Once the session is active, you can continue the conversation normally without repeating @agent on every turn.

To end the session early, use:

/exit

💡

Agent sessions are interactive loops, not one-shot prompts. RealTimeX will hand the chat over to the agent until the session completes or you end it with /exit.

What agents can use

The current RealTimeX agent model is broader than the older built-in tool list. Depending on how your workspace is configured, an agent can use:

workspace documents and search context
workspace system prompts
workspace personality files
built-in and imported skills
agent flows
remote or local MCP servers
configured Agentic CLIs
working directories and writable wiki projects
browser sessions through Browser Tool and Agent Browser
stored credentials used by advanced workflows
runtime-injected provider authentication when supported by the launched agent

Where agent capability comes from

Workspace Agent configuration

Use Workspace Settings > Agent to control the main capability layer for that workspace.

The current tabs are:

Skills: built-in plugin skills and imported skills
Agent Flows: reusable flow automations
MCP Servers: remote and local MCP integrations

Runtime controls

Use Agent Runtime for the operational side of agents:

Agentic CLIs
working directories
Ambient Agent
diagnostics and agent feed

Plugins

Use Plugins when the capability depends on a built-in or installed plugin, such as:

Typical use cases

search workspace material and summarize findings
work through browser-backed research or QA tasks
run flow-based automations from chat
call external tools through MCP servers
work across trusted folders or wiki projects
produce exports, charts, or structured outputs when the right skills are enabled

Good prompts

Examples that fit the current product well:

@agent review the documents in this workspace and give me a decision brief
@agent use the browser to compare these three pricing pages
@agent update the wiki project in my working directory with today's findings
@agent use the enabled MCP tools to check the deployment status

What to configure first when an agent feels limited

If the agent cannot do what you expect, check these layers in order:

Workspace Settings > Agent
Agent Runtime
Plugins
Credentials, Agent Authentication, or behavior config such as System Prompts and Personality

All Features API Access