Inputs and Priorities

The crate splits inputs into eager text records and lazy file records. Priority influences which chunks survive when the token budget tightens.

Content inputs#

rust

use forgetless::{FileWithPriority, Forgetless, Priority, WithPriority};

let png = std::fs::read("diagram.png")?;

let result = Forgetless::new()
    .add("plain medium-priority text")
    .add(WithPriority::high("recent discussion"))
    .add_pinned("system instructions")
    .add_file("notes.md")
    .add_file(FileWithPriority::critical("architecture.pdf"))
    .add_bytes(&png, "image/png")
    .add_bytes_p(&png, "image/png", Priority::High)
    .run()
    .await?;

Priority helpers#

Helper	Priority	Where it lives
WithPriority::critical	Critical	Text content wrapper for `.add()`.
WithPriority::high	High	Text content wrapper for `.add()`.
WithPriority::medium	Medium	Explicit default if you want to be verbose.
WithPriority::low	Low	Background context with weaker retention.
FileWithPriority::critical	Critical	Lazy file wrapper for `.add_file()`.
FileWithPriority::high	High	Lazy file wrapper for `.add_file()`.
FileWithPriority::medium	Medium	Explicit medium file priority.
FileWithPriority::low	Low	Lazy file wrapper for lower-value file context.

Lazy files#

Forgetless stores file paths first and delays real reads until run(). That matters because the builder can preview and filter paths before paying the cost of parsing every PDF, image, or source tree.

Behavior	Implementation detail
Default priority	Plain path inputs become `LazyFile` with `Priority::Medium`.
Accepted path forms	`&str`, `String`, `PathBuf`, `&Path`, and `FileWithPriority<P>`.
Preview phase	When a query exists and more than five lazy files were registered, previews are used to shortlist files.
Read phase	Only shortlisted files are converted into `ContentInput` and chunked.

Pipeline

See how the Forgetless context optimization pipeline runs: preview filtering, parallel file reads, chunking, relevance scoring, and budget-based selection.

Chunking

Configure content-aware chunking for LLMs: pick presets by content type and tune deduplication, overlap, and target token size for accurate ranking.

Introduction

Learn how Forgetless, a Rust LLM context optimizer, ranks and compresses oversized prompts and files into budget-aware context for production workflows.