fix: make compression more aggressive

fix: compress user and tagged messages by default

small redis fix

Require explicit Redis env for aggregation

Add live Redis multi-worker integration tests

Support Redis env alias for multi-worker stats

fix: remove legacy HEADROOM_REDIS_URL fallback, use REDIS_URL only

feat(proxy): Upstash Redis HTTP REST support for HF Spaces multi-worker stats

fix(tokenizers): route Claude models to tiktoken cl100k_base

refactor: use tiktoken universally for all model token counting

fix: route count_tokens to api.anthropic.com, not proxy base_url

chore(rust): SmartCrusher CCR marker injection + walker unification

chore(rust): SmartCrusher CCR storage layer + roundtrip verification

feat(rust): SmartCrusher PR4 — lossless-first default + CCR-Dropped restoration

feat(rust): retire python smart_crusher, ship rust-only via pyo3

feat(rust): pyo3 bridge for SmartCrusher

fix(python+rust): smart_crusher bugs #1, #2, #3, #4 + sorted iteration

feat(rust): retire python diff_compressor, ship rust-only via pyo3

fix(diff): close ContentRouter routing gaps for merge diffs and long preambles

fix(diff_compressor): four silent information-loss paths in Python AND Rust

Update number array tests for schema-preserving compression

Fix SmartCrusher bugs: schema violation, race condition, thread safety, recursion

Make KompressCompressor model-configurable: model_id, chunk_words, score_threshold

feat(kompress): add compress_batch with device-aware routing

Fix tests: update default assertions for disabled CodeCompressor

Kompress ONNX INT8: text compression without torch dependency

Fix CI test, bump to 0.5.7

Remove LLMLingua: Kompress is the sole text compressor

Protect workflow XML tags from text compression

Diversity-aware SmartCrusher: keep unique items, compress text within

Fixing tests

Fix proxy crash when torch not installed (kompress lazy imports)

Add Kompress: ModernBERT token compressor replacing LLMLingua-2

feat: provider-aware prefix cache tracking and combined savings dashboard

fix: dashboard metrics, TTFB tracking, eager LLMLingua loading, and multi-provider consistency

feat(router): adaptive compression with Read lifecycle and context-pressure scaling

feat(code): add semantic symbol importance to CodeAwareCompressor

Add Read Lifecycle: event-driven stale/superseded Read detection

Add OSS evaluation suite, universal JSON crush, latency benchmarks

test: Update LLMLingua default model test to match new default

fix: Extend Anthropic format tool protection to IntelligentContextManager

fix(tests): add missing skip decorator and tests for exclude_tools

fix(tests): skip HTML extractor tests when trafilatura not installed

Add HTMLExtractor for web content extraction with OSS benchmarks

Add DiffCompressor and fix hnswlib SIGILL crash on CI

feat(toin): add TOIN + CCR integration to IntelligentContextManager

Remove hardcoded source hint system from ContentRouter

Add comprehensive evaluation framework for compression accuracy

Add reasoning capability forwarding to HeadroomAgnoModel

Add dynamic anchor selection with content deduplication