Struct TokenBufferMemory 

pub struct TokenBufferMemory { /* private fields */ }

Conversation memory that keeps the most recent messages that fit in max_tokens.

When a new message pushes the buffer over the limit, the oldest messages are dropped until it fits again. A tool result left at the front without the AI message that requested it is dropped as well, so the history stays valid for chat APIs. A single message bigger than max_tokens is not kept.

Tokens are counted with tiktoken (cl100k_base by default) over each message’s content plus its tool calls, which is an approximation for non-OpenAI models.

let memory = TokenBufferMemory::new(2000).with_model("gpt-4o")?;

Implementations

impl TokenBufferMemory

pub fn new(max_tokens: usize) -> Self

pub fn with_model(self, model: &str) -> Result<Self, MemoryError>

Count tokens with the tokenizer tiktoken uses for model (e.g. gpt-4o).

pub fn with_token_counter<F>(self, counter: F) -> Self

where F: Fn(&str) -> usize + Send + Sync + 'static,

Use a custom token counter, e.g. one matching a non-OpenAI model’s tokenizer.

pub fn total_tokens(&self) -> usize

Tokens currently held in the buffer.

Trait Implementations

impl BaseMemory for TokenBufferMemory

fn messages<'life0, 'async_trait>( &'life0 self, ) -> Pin<Box<dyn Future<Output = Result<Vec<Message>, MemoryError>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait,

fn add_message<'life0, 'async_trait>( &'life0 mut self, message: Message, ) -> Pin<Box<dyn Future<Output = Result<(), MemoryError>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait,

fn clear<'life0, 'async_trait>( &'life0 mut self, ) -> Pin<Box<dyn Future<Output = Result<(), MemoryError>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait,

fn add_user_message<'life0, 'life1, 'async_trait>( &'life0 mut self, message: &'life1 (dyn Display + Sync), ) -> Pin<Box<dyn Future<Output = Result<(), MemoryError>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait,

fn add_ai_message<'life0, 'life1, 'async_trait>( &'life0 mut self, message: &'life1 (dyn Display + Sync), ) -> Pin<Box<dyn Future<Output = Result<(), MemoryError>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait,

fn to_string<'life0, 'async_trait>( &'life0 self, ) -> Pin<Box<dyn Future<Output = Result<String, MemoryError>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait,

impl From<TokenBufferMemory> for Arc<dyn BaseMemory>

fn from(memory: TokenBufferMemory) -> Self

Converts to this type from the input type.

impl From<TokenBufferMemory> for Arc<Mutex<dyn BaseMemory>>

fn from(memory: TokenBufferMemory) -> Self

Converts to this type from the input type.