Agents
An agent is a reusable role configuration. Choosing one when creating a session brings in its model, prompts, tools, MCP servers and skills.
Create an agent
Section titled “Create an agent”Go to Agents → New agent config in the left menu.
Basics
Section titled “Basics”| Field | Notes |
|---|---|
| Config name | Shown when picking an agent; name it by purpose, e.g. “code review” |
| Role label | Free text, used only for display and filtering |
- Bind LLM config chooses the model for this agent.
- Follow system default uses the current default model; changing the default updates every agent that follows it.
Prompts
Section titled “Prompts”- System prompts define role, boundaries and output style. You can select several; they are concatenated in order.
- User prompts act as user-side openings or constraints, also multi-select.
Prompts are maintained under Prompts; an agent only selects them.
Resources
Section titled “Resources”- Built-in tools — tick the tools this agent may use.
- MCP servers — tick the external tool servers to attach.
- Skills — grouped selection; search by name when the list is long.
All three support multi-select. Leaving them empty means the agent uses none of that capability.
Use an agent
Section titled “Use an agent”Pick it in the agent dropdown when creating a session. Depending on the UI you can also switch inside an existing session.
Edit and delete
Section titled “Edit and delete”- Edits apply from the next turn.
- Deleting only affects future selections. Make sure no running session depends on it first.
The agent does not use a tool I expected. Check it is ticked, then check the tool’s timeout and output truncation settings.
I changed the default model but the agent did not change. It is bound to a specific LLM config. Switch it to “follow system default”.
Prompts are too long. Keep stable rules in system prompts and put per-task instructions in the message.