Module 4 · Conversations & Chat

Anatomy of a Conversation

Every conversation is a self-contained context with its own model, system prompt, think level, and message history. Key fields:

Field	Purpose
Title	Display name (auto-generated or user-set)
ModelName	Which model processes this conversation
SystemPrompt	Persona / instructions prepended to every request
ThinkLevel	Reasoning depth (see below)
Messages	Ordered list of ChatMessage objects (cascade-deleted with the conversation)
IsArchived	Hides the conversation without deleting it

Think Levels

Think levels are the most powerful knob in DaisiBot. They control how deeply the model reasons before responding.

Level	Behavior	Tools?
Basic	Direct inference — fast, simple answers.	No
BasicWithTools	Direct inference with access to enabled tool groups.	Yes
Skilled	Single-pass reasoning with visible thinking. The model reasons through the entire problem in one inference call, "thinking out loud" as it works toward an answer. Good for complex questions that benefit from visible reasoning.	Yes
Agent	Autonomous multi-step execution. Instead of answering in a single pass, the model first plans a set of steps, then executes each step individually (with full tool access), and finally synthesizes all results into a single response. This plan–execute–synthesize loop lets the model break down complex tasks and tackle them methodically.	Yes

Not every model supports every level. The AvailableModel object exposes a SupportedThinkLevels list. Both Skilled and Agent work with any model that supports reasoning, though Agent benefits from models that produce structured plans.

Streaming Responses

DaisiBot streams every response as an IAsyncEnumerable<StreamChunk>. Each chunk carries:

• Content – The text fragment.
• Type – Text, Thinking, Tooling, ToolContent, Error, Image, or Audio.
• IsComplete – true on the final chunk.

The UI renders text chunks progressively (typewriter effect) and separates Thinking chunks into a collapsible section so you can see the model's chain-of-thought without cluttering the response.

Message Types

Each ChatMessage has a Role and a Type:

Slash Commands (TUI)

In the terminal client, type / to open the autocomplete popup. The top 5 matches appear and you navigate with Up/Down, complete with Tab or Enter.

Command	What It Does
/new	Create a new conversation
/model	Open the model picker
/settings	Open the settings dialog
/skills	Open the skill browser
/install <skill>	Install a skill by name
/login	Start the OTP authentication flow
/balance	Check your DAISI credit balance
/status	Show bot status or all bots
/runnow	Run a selected bot immediately
/kill	Stop and delete a bot (with confirmation)
/export	Export the current conversation
/clear	Clear the screen
/help	Show help information

Chat Statistics

After every response, DaisiBot tracks token count, compute time, and tokens per second for both the last message and the running session total. In the TUI these appear in the ChatStatsBar; in MAUI/Web they appear below each message bubble. Use these to compare model performance and estimate costs.

Active Recall Checkpoint

Cover the answers and test yourself before peeking.