Crate ainl_compression 

Heuristic compression primitives for AINL hosts.

This crate currently exposes the ArmaraOS “Ultra Cost-Efficient Mode” input compressor. It is intentionally embedding-free and dependency-light so it can be reused across hosts without shipping local ML models.

Companion modules (no I/O; host wiring stays out-of-crate):

• profiles — built-in compression profiles and project→profile hints
• adaptive — content-shaped EfficientMode recommendations
• cache — TTL hysteresis for cache-aware coordination

Set RUST_LOG=ainl_compression=debug to enable full before/after text logging per call (useful for tuning preserve lists and retention ratios).

Re-exports

pub use adaptive::recommend_mode_for_content;

pub use adaptive::AdaptiveRecommendation;

pub use cache::cache_policy_summary;

pub use cache::effective_ttl_with_hysteresis;

pub use cache::CacheTtlResult;

pub use profiles::list_builtin_profiles;

pub use profiles::resolve_builtin_profile;

pub use profiles::suggest_profile_id_for_project;

pub use profiles::CompressionProfile;

pub use profiles::BUILTIN_PROFILES;

Modules

adaptive

Embedding-free content classifier for adaptive eco mode selection.

cache

Cache-aware TTL coordination hints for compression-related caches.

profiles

Built-in compression profiles (policy vocabulary for hosts).

Structs

Compressed

Result of a compression pass.

CompressionMetrics

Structured telemetry emitted for each compression operation.

PromptCompressor

Standalone input prompt compressor.

Enums

EfficientMode

Input compression aggressiveness.

Traits

CompressionTelemetrySink

Optional telemetry sink for compression metrics.

Functions

compress

Compress text toward mode.retain() of its original token budget.

compress_with_metrics

Compress and return structured telemetry metrics in one call.

estimate_semantic_preservation_score

Lightweight lexical semantic-preservation heuristic in range 0.0..1.0.

tokenize_estimate

Heuristic token estimate (~4 chars per token, +1 minimum).