Struct ProgressiveTransformer 

pub struct ProgressiveTransformer { /* private fields */ }

A two-phase output transformer that returns an LLM-generated summary first, with the full output cached and available on demand via the __get_full_output virtual tool.

Workflow

1. Output arrives from a tool call.
2. If output.len() <= max_summary_length → returned unchanged (no LLM call).
3. If larger → LLM is called to summarize → summary returned + note appended.
4. Full output cached internally keyed by tool_name.
5. Agent can call __get_full_output → FullOutputRetriever serves it.

Example

use traitclaw_core::transformers::ProgressiveTransformer;
use std::sync::Arc;

// let transformer = ProgressiveTransformer::new(provider.clone(), 500)
//     .with_summary_prompt("Give a one-sentence summary: {output}");

Implementations

impl ProgressiveTransformer

pub fn new(provider: Arc<dyn Provider>, max_summary_length: usize) -> Self

Create a new progressive transformer.

• provider: LLM used to generate summaries.
• max_summary_length: Outputs shorter than this are passed through unchanged.

pub fn with_summary_prompt(self, prompt: impl Into<String>) -> Self

Override the summarization prompt template.

Use {output} as the placeholder for the tool output.

pub fn retriever_tool(&self) -> FullOutputRetriever

Build a FullOutputRetriever that reads from this transformer’s cache.

Register this tool with the agent so the LLM can call __get_full_output.

Trait Implementations

impl OutputTransformer for ProgressiveTransformer

fn transform<'life0, 'life1, 'life2, 'async_trait>( &'life0 self, output: String, tool_name: &'life1 str, _state: &'life2 AgentState, ) -> Pin<Box<dyn Future<Output = String> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait, 'life2: 'async_trait,

Transform tool output, optionally using context about which tool produced it and the current agent state.

fn estimate_output_tokens(&self, output: &str) -> usize

Estimate token count for a given output string.