Agents

An Agent is a named assistant configured for a particular kind of work. Every Agent inherits one global capability configuration and stores only its profile, workspace choice, and explicit overrides. User-owned context remains available across Agents according to your settings.

When to create another Agent

Create one when you need a real capability boundary, for example:

• a coding Agent allowed to edit repositories;
• a research Agent with web and connector access;
• a personal Agent that uses a separate workspace;
• a lightweight Agent that uses a lower-cost model.

Do not create a new Agent only to start a different conversation. A new Session is enough for that.

Create an Agent in the console

1. Open Agents.
2. Choose Add Agent.
3. Give it a clear name and personality instruction.
4. Save, then start a new chat with that Agent.
5. Add an Agent-specific override only when it must differ from the global defaults.

Test a read-only request before allowing tools that write files, run commands, send messages, or access external accounts.

Create an Agent in the terminal

xopc agents add research
xopc agents list

The command creates or updates the Agent entry and prepares its directories. To remove an Agent from configuration:

xopc agents delete research

Read the confirmation carefully if on-disk cleanup is offered; deleting files is different from disabling an Agent.

Choose the default Agent

The default is used for new Sessions that do not name an Agent:

xopc config set agents.default research
xopc config validate

Existing Sessions remain assigned to their current Agent. To use another Agent, start a new chat and select it explicitly.

Design a safe Agent

For each Agent, decide:

Setting	Question to answer
Responsibility	What work should this Agent accept or decline?
Workspace	Which files may it use?
Models	Which model balances quality, speed, privacy, and cost?
Tools	Which actions can it take?
Skills	Which reusable instructions does it need?
Boundaries	Which actions always require confirmation?

Begin with the smallest useful capability set. Add access only after the Agent succeeds without it.

For a guided example, see Create a second Agent. Exact configuration fields are listed in Configuration reference.