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

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

style: ruff format parser.py, content_router.py, smart_crusher.py

fix: harden edge cases, expand tool exclusions, and fix mypy errors

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Add Read Lifecycle: event-driven stale/superseded Read detection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

Add one-function compress() API, ASGI middleware, LiteLLM callback

Add Compression Hooks — extension points for SaaS and advanced customization

Add Query Echo: re-inject user question after compressed tool outputs

Add compression summaries, multi-provider headers, Dockerfile fix

Add adaptive compression sizing and fix cost tracking bugs

Add LiteLLM backend routing for OpenAI endpoint and Magika content detection

Add centralized ML model configuration

fix: Extend Anthropic format tool protection to IntelligentContextManager

fix: Handle Anthropic format tool_use/tool_result as atomic units

Merge remote-tracking branch 'origin/main'

Add question-aware compression and enable ContentRouter by default

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

feat(compression): add exclude_tools to bypass compression for specific tools

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

feat(toin): extend TOIN integration to all ContentRouter compression strategies

Add image token compression with trained ML router

Update compressor CCR markers and add batch compression eval

Remove hardcoded source hint system from ContentRouter

Add comprehensive evaluation framework for compression accuracy

Fix TOIN integration and add persistence support

Add reasoning capability forwarding to HeadroomAgnoModel

Add dynamic anchor selection with content deduplication

Integrate DynamicContentDetector into CacheAligner (Phase 1)

Add quality retention eval and fix linting for Python 3.12

Add Phase 2 Progressive Summarization and fix Agno integration tests

Implement COMPRESS_FIRST strategy for IntelligentContextManager

Add IntelligentContextManager for semantic-aware context management

Fix mypy type errors for Python 3.12 compatibility

Fix ruff formatting

Add TOIN field-level learning and comprehensive integration tests

Fix mypy type annotation error in content_router.py

Add CCR marker support to all compressors and fix retrieval

Fix LLMLingua API: use context=[content] not original_prompt

Fix 'list' object has no attribute 'strip' error

Add seamless LangChain integration

Add AST-based code compression and custom model configuration

Add LLMLingua-2 opt-in support to proxy server

Fix mypy errors: correct CompressionStore.store() arguments

Fix lint errors in text compression utilities

Add opt-in text compression utilities for coding tasks