Struct Brain 

pub struct Brain { /* private fields */ }

The Brain wraps an LLM provider and adds higher-level logic: prompt construction, cost tracking, and model selection.

Implementations

impl Brain

pub fn new( provider: Arc<dyn LlmProvider>, system_prompt: impl Into<String>, ) -> Self

pub fn set_knowledge_addendum(&mut self, addendum: String)

Set knowledge addendum (distilled rules) to append to the system prompt.

pub fn estimate_tokens(&self, messages: &[Message]) -> usize

Estimate token count for messages using tiktoken-rs.

pub fn estimate_tokens_with_tools( &self, messages: &[Message], tools: Option<&[ToolDefinition]>, ) -> usize

Estimate token count for messages plus tool definitions.

pub fn build_messages(&self, conversation: &[Message]) -> Vec<Message>

Construct messages for the LLM with system prompt prepended.

If a knowledge addendum has been set via set_knowledge_addendum(), it is automatically appended to the system prompt.

After assembly, sanitize_tool_sequence runs to ensure tool_call→tool_result ordering is never broken regardless of compression, pinning, or system message injection.

pub async fn think( &mut self, conversation: &[Message], tools: Option<Vec<ToolDefinition>>, ) -> Result<CompletionResponse, LlmError>

Send a completion request and return the response, tracking usage.

pub async fn think_with_retry( &mut self, conversation: &[Message], tools: Option<Vec<ToolDefinition>>, max_retries: usize, ) -> Result<CompletionResponse, LlmError>

Send a completion request with retry logic and exponential backoff.

Retries on transient errors (RateLimited, Timeout, Connection) up to max_retries times with exponential backoff (1s, 2s, 4s, …, capped at 32s). Non-transient errors are returned immediately.

pub fn is_retryable(error: &LlmError) -> bool

Check if an LLM error is transient and should be retried.

pub async fn think_streaming( &mut self, conversation: &[Message], tools: Option<Vec<ToolDefinition>>, tx: Sender<StreamEvent>, ) -> Result<(), LlmError>

Send a streaming completion request, returning events via channel.

pub fn total_usage(&self) -> &TokenUsage

Get total token usage across all calls.

pub fn total_cost(&self) -> &CostEstimate

Get total cost across all calls.

pub fn model_name(&self) -> &str

Get the model name.

pub fn context_window(&self) -> usize

Get the context window size.

pub fn provider_cost_rates(&self) -> (f64, f64)

Get cost rates (input_per_token, output_per_token) from the provider.

pub fn provider(&self) -> &dyn LlmProvider

Get a reference to the underlying LLM provider.

pub fn provider_arc(&self) -> Arc<dyn LlmProvider>

Get a cloneable Arc handle to the underlying LLM provider.

pub fn track_usage(&mut self, usage: &TokenUsage)

Track usage and cost from an external completion (e.g., streaming).

pub fn context_usage_ratio(&self, conversation: &[Message]) -> f32

Get the current token usage as a fraction of the context window.