Skip to content

Changelog

All notable changes to StrataRouter will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Unreleased]

Added

  • Multi-model routing support in runtime executor
  • Advanced policy engine with TCFP workflows
  • Distributed tracing with OpenTelemetry
  • Enhanced semantic caching with similarity search
  • Batch processing with intelligent deduplication

Changed

  • Improved HNSW index performance with SIMD optimization
  • Enhanced error handling in core-runtime bridge
  • Updated Python bindings for better type safety

Fixed

  • Memory leak in cache invalidation
  • Race condition in batch coordinator
  • Edge cases in confidence calibration

[0.3.0] - 2025-01-10

Added

  • Core-Runtime Bridge: Production-grade integration layer
  • Routing Executor: Multi-provider execution engine
  • State Management: Transaction-safe state operations
  • Observability: Comprehensive metrics and distributed tracing
  • Caching Layer: Semantic and similarity-based caching
  • Batch Processing: Request batching and deduplication
  • Rate Limiting: Token bucket and leaky bucket algorithms
  • Sandbox Execution: Isolated execution environment
  • Feedback Loop: Learning from routing outcomes

Changed

  • Refactored runtime architecture for better modularity
  • Improved hybrid scoring algorithm accuracy (95.4% → 96.1%)
  • Enhanced HNSW index with dynamic graph updates
  • Optimized memory footprint (68MB → 64MB base)

Fixed

  • SIMD operations on non-AVX2 systems
  • Thread safety in concurrent index updates
  • Edge cases in isotonic regression calibration
  • Memory alignment issues in vector operations

[0.2.0] - 2024-12-15

Added

  • Hybrid Scoring: Semantic + keyword + rule-based routing
  • Confidence Calibration: Isotonic regression for accurate probabilities
  • SIMD Optimization: AVX2 vectorization for 10x performance boost
  • Python Bindings: Complete PyO3-based Python API
  • Framework Integrations: LangChain, LangGraph, CrewAI, AutoGen
  • Benchmarking Suite: Comprehensive performance testing
  • Documentation: Full API reference and guides

Changed

  • Improved HNSW index construction (2x faster)
  • Enhanced route matching with multiple strategies
  • Better error messages with context

Fixed

  • Index corruption on concurrent updates
  • Memory leaks in embedding cache
  • Incorrect confidence scores edge cases

Performance

  • P99 latency: 12ms → 8.7ms
  • Throughput: 12K → 18K requests/second
  • Memory: 82MB → 68MB base footprint

[0.1.0] - 2024-11-20

Added

  • Initial release
  • HNSW Index: Fast approximate nearest neighbor search
  • Basic Routing: Semantic similarity-based routing
  • Route Management: Add, update, delete routes
  • Embeddings: Support for multiple embedding models
  • Configuration: YAML and Python configuration
  • CLI: Basic command-line interface
  • Testing: Unit and integration tests

Features

  • Sub-20ms P99 latency
  • 10K+ requests/second throughput
  • ~100MB memory footprint
  • Thread-safe operations
  • OpenAI and local model support

Version History

Core Engine

Version Release Date Key Features
0.3.0 2025-01-10 Runtime integration, advanced features
0.2.0 2024-12-15 Hybrid scoring, SIMD optimization
0.1.0 2024-11-20 Initial release, HNSW index

Runtime

Version Release Date Key Features
0.3.0 2025-01-10 Complete runtime with execution engine
0.2.0-beta 2024-12-20 Beta release for testing

Migration Guides

Migrating to 0.3.0

Breaking Changes:

  • Runtime configuration moved to separate file
  • Some API methods renamed for consistency
  • Python bindings require stratarouter[runtime] extra

Migration Steps:

# Old (0.2.x)
from stratarouter import Router
router = Router()

# New (0.3.x)
from stratarouter import Router
from stratarouter_runtime import RoutingExecutor

router = Router()
executor = RoutingExecutor(router)

Configuration:

# Old
router:
  embedding_dim: 384

# New
core:
  embedding_dim: 384
runtime:
  timeout_ms: 30000

Migrating to 0.2.0

Breaking Changes:

  • route_query() now returns RouteResult instead of string
  • HNSW parameters renamed for clarity
  • Python API requires type hints

Migration Steps:

# Old (0.1.x)
result = router.route_query("query text")
route_id = result

# New (0.2.x)
result = router.route_query("query text", embedding)
route_id = result.route_id
confidence = result.confidence

Deprecations

Deprecated in 0.3.0

  • Router.route_simple() - Use Router.route() with default params
  • RouterConfig.use_gpu - GPU support moved to embedding models

Removed in 0.3.0

  • Router.add_route_batch() - Use Router.add_routes()
  • Legacy configuration format

Performance Improvements

Version 0.3.0

  • 15% faster route matching with optimized HNSW search
  • 30% memory reduction with sparse embeddings
  • 2x faster batch processing with deduplication

Version 0.2.0

  • 10x faster with SIMD (AVX2) optimization
  • 45% faster HNSW construction
  • 40% less memory with optimized storage

Version 0.1.0

  • Baseline performance metrics established

Security Updates

0.3.0

  • Added request validation and sanitization
  • Implemented rate limiting to prevent abuse
  • Enhanced sandbox execution security
  • Added API key authentication support

0.2.0

  • Fixed potential XSS in error messages
  • Improved input validation
  • Added secure defaults for all configs

Known Issues

Current (0.3.0)

  • High memory usage with >10K routes (optimization in progress)
  • Occasional timeout with very long queries (>1000 tokens)
  • Limited GPU support (planned for 0.4.0)

Fixed in 0.3.0

  • Memory leak in cache invalidation
  • Race condition in batch processing
  • SIMD crashes on non-AVX2 systems

Roadmap

See Roadmap for future plans.

Planned for 0.4.0

  • GPU acceleration for embedding generation
  • Advanced multi-modal routing (text + images)
  • Streaming support for long-running queries
  • Enhanced policy engine with custom rules
  • Distributed deployment support

Under Consideration

  • gRPC API for better performance
  • WebAssembly builds for edge deployment
  • Advanced analytics and insights
  • Auto-scaling based on load

Contributors

Thank you to all our contributors! See GitHub Contributors for the full list.

Support

License

StrataRouter is licensed under the MIT License. See LICENSE for details.

Note: This changelog follows Keep a Changelog format. For more details on any release, see the corresponding GitHub Release.