Conversations
What a conversation is for
A conversation is the main interaction space in Kitemesh. It stores message history, attachments, generated output, and any tool approval steps that may need user action.
Start a new chat
To start a new chat:
- select
New Chat - choose an agent or group
- type the message
- send the request
The selected target determines the behavior of the chat, the visible resources, and the available tools.
Operating modes
Depending on the configuration, several modes can be available:
AutoNormalOrchestrationSynthesis
In practice:
Normalfits simple requestsOrchestrationhelps with multi-step workSynthesisworks well when several agents contribute to one responseAutolets Kitemesh choose
Attachments and context
From the chat configuration panel, it is possible to:
- include or exclude specific attachments
- inspect where those attachments come from
- adjust the tools that should be included or excluded
- exclude API resources for the current conversation
- exclude MCP tools for the current conversation
This keeps the working context clean and prevents a chat from using irrelevant files or tools.
Temporary chat
Temporary chat is available only before a new conversation starts. Use it when the conversation should stay separate from the normal history.
If the option is disabled, open a new chat before enabling it.
Chat settings
Depending on the selected model, Chat settings can expose:
Brain temperatureReasoning effortContext flowOrchestration mode
Some models do not support every setting. When a setting is marked as unsupported, Kitemesh keeps the model or provider default.
Live reasoning
Some flows can display Live Reasoning.
This view helps readers:
- follow the current processing steps
- see which agent is working
- understand whether orchestration is running, completed, or interrupted
Conversation history and actions
From the conversations panel, it is possible to:
- reopen an existing conversation
- rename it
- delete it
- load more history
When to start a new conversation
Starting a new chat is usually the better option when:
- the goal changes
- the selected group model changes
- the selected custom model changes
- the file context needs to be isolated
- the previous history should not be reused