Struct ConversationHandler 

pub struct ConversationHandler { /* private fields */ }

Conversation handler for multi-turn LLM interactions

Implementations

impl ConversationHandler

pub fn new( system_message: String, client: Arc<OllamaClient>, model: String, rate_limiter: RateLimiter, request_source: RequestSource, ) -> Self

Create a new conversation handler with a system message

pub fn with_bridge_event( self, server_id: ServerId, connection_id: Option<ConnectionId>, protocol: &str, event: &Event, ) -> Self

Attach the event for a bridge host without making it recover context from prompt text.

pub fn with_history_limits(self, max_messages: usize, max_chars: usize) -> Self

Override the conversation-history caps (see DEFAULT_MAX_HISTORY_MESSAGES and DEFAULT_MAX_HISTORY_CHARS).

pub fn trim_history(&mut self)

Bound the conversation that is actually sent to the model.

Without this, messages only ever grows: five tool iterations each append an assistant response, an acknowledgement and a tool-results block, and every failed attempt leaves its response plus a correction behind for the rest of the conversation. ConversationState’s 8000-char window is a different structure and bounds only cross-call history, never the request.

The system message and the first user message — the instruction and the request the whole conversation is about — are never dropped, nor are the two most recent messages. Everything between is dropped oldest-first and replaced by a single [TRIM_NOTICE], so the model is told its history was cut rather than left to reference turns that silently vanished.

Called before every request; public so the bound is directly assertable.

pub fn with_private_payloads(self, private: bool) -> Self

Hide incidental payload/error diagnostics when structured request input contains credentials. Sticky only within this conversation’s history.

pub fn with_status_tx(self, tx: UnboundedSender<String>) -> Self

Set the status channel for user-visible logs

pub fn with_conversation_state( self, state: Arc<Mutex<ConversationState>>, ) -> Self

Set an existing conversation state

pub fn with_tracking( self, state: AppState, source: ConversationSource, details: String, ) -> Self

Set conversation tracking information

pub fn with_native_tools(self, actions: &[ActionDefinition]) -> Self

Enable native tool calling with the given action definitions

When set, generate_with_retry() will use chat_with_tools() instead of generate_with_format(), sending structured messages and tool schemas. The response is converted back to ActionResponse text format so the existing tool calling loop in generate_with_tools_and_retry() works unchanged.

Also strips redundant action/tool descriptions and JSON format instructions from the system message to reduce token usage (these are now handled by native tool schemas).

pub fn update_tool_schemas(&mut self, actions: &[ActionDefinition])

Update the native tool schemas (e.g., after documentation is read and new tools are enabled)

pub async fn end_tracking(&self)

Manually end conversation tracking (for error paths where generate_with_tools_and_retry doesn’t complete)

pub fn is_protocol_docs_read(&self) -> bool

Check if protocol documentation has been read in this conversation

pub fn add_user_message(&mut self, content: String)

Add a user message to the conversation

pub async fn generate_with_tools_and_retry( &mut self, approval_tx: Option<UnboundedSender<WebApprovalRequest>>, web_search_mode: WebSearchMode, available_actions: Vec<ActionDefinition>, ) -> Result<Vec<Value>>

Generate response with tool calling and retry logic

This is the main entry point that handles:

1. Multi-turn tool calling loop
2. Automatic retry on parse/validation errors
3. Tool execution with result feedback

Arguments

• approval_tx - Optional channel for web search approval
• web_search_mode - Web search configuration
• available_actions - List of actions available in this context

Returns

• Ok(Vec<serde_json::Value>) - Array of non-tool actions to execute

pub fn messages(&self) -> &[Message]

Get the current conversation messages (for debugging)

pub fn message_count(&self) -> usize

Get the number of messages in the conversation

pub fn conversation_id(&self) -> &str

Get the conversation ID

pub fn mark_registered(&mut self)

Mark this conversation as already registered (to prevent duplicate registration)

pub async fn update_current_state( &mut self, state: &AppState, server_id: Option<ServerId>, )

Update the “Current State” section in the system message

This is used after actions like open_server that modify application state. The system message is rebuilt with updated state so subsequent tool calls see the current state.

Arguments

• state - Application state
• server_id - Optional server context