Group Models
What a group model is for
A group model assembles several components into one chat experience. It is the main entry point selected before sending a request.
Main components
A group model can contain:
- one
brain - zero, one, or several
agents - a default mode
- a category, tags, and an icon
- visibility and shared teams
- optional resources and tools shared with the group
Model references
The Brain and each agent can use:
- an LLM model
- a custom model
- another group model
Avoid circular group references. If a group cannot be saved, check whether one group points back to another group that already contains it.
The role of the brain
The brain is the main model in the group. It carries the overall coordination and, depending on the selected mode, can distribute work to the agents.
The role of the agents
Each agent can:
- have a role
- receive an instruction override
- focus on a specific part of the task
The goal is not to add many agents by default, but to separate responsibilities clearly when needed.
Default mode
The default mode defines how the group behaves when a new chat starts. A typical choice depends on the dominant need:
Autofor flexibilityNormalfor one direct answerOrchestrationfor complex tasksSynthesisfor combining several contributions
Temperature
When a selected model supports it, the group can define a custom brain or agent temperature. If temperature is not supported, Kitemesh keeps the model or provider default.
Group resources and tools
The group detail page includes Tools and group files. Use them when the tool or resource should be available whenever this group is selected.
When to create a new group
A dedicated group is usually the right choice when:
- the main use case changes
- a different set of agents is needed
- the coordination logic needs to change
- resources or access rules should be separated
Good practices
- Use a name that describes the purpose clearly.
- Start with a small number of agents.
- Avoid overly broad groups.
- Review sharing before making a group public.