Structs§
- Condenser
Middleware - Middleware that applies a condenser to messages before each model call.
- LlmSummarizing
Condenser - Summarizes older messages using an LLM, keeping recent messages intact.
- NoOp
Condenser - A no-op condenser that returns messages unchanged.
- Pipeline
Condenser - Chains multiple condensers in sequence.
- Rolling
Condenser - Keeps only the most recent N messages, optionally preserving the system message.
- Token
Budget Condenser - Trims messages to fit within a token budget, keeping the most recent messages.
Traits§
- Condenser
- Trait for condensing (compressing) a message history.