Skip to content

Status: Needs Review

This page has not been reviewed for accuracy and completeness. Content may be outdated or contain errors.

Changelog

This page provides a high-level overview of cuvis-ai releases. For the complete changelog with all technical details, see the main CHANGELOG.md.

Latest Release

Version 0.2.4 (February 5, 2026)

Documentation Complete! This release marks the completion of comprehensive documentation for cuvis-ai, making the framework more accessible and easier to use for everyone.

Highlights:

  • 70+ Documentation Pages covering all aspects of the framework
  • 6 Step-by-Step Tutorials for common workflows (RX Statistical, Channel Selector, Deep SVDD, AdaCLIP, gRPC, Remote Deployment)
  • Complete Node Catalog documenting 50+ built-in nodes across 11 categories
  • Plugin System Guides for extending the framework with custom nodes
  • gRPC API Documentation with sequence diagrams and client patterns
  • Configuration References for Hydra composition and schema definitions
  • 95%+ Docstring Coverage across all public APIs
  • Central Plugin Registry for community plugin discovery

New Documentation:

  • API reference with auto-generated documentation from docstrings
  • 7 how-to guides for building pipelines, restoration, monitoring, and more
  • Development guides for contributing and maintaining code quality
  • 20+ diagrams (Mermaid + Graphviz) for visual understanding

Improvements:

  • README refactored to eliminate duplicate content
  • Contributing guide enhanced with complete plugin contribution workflow
  • All code examples updated to use uv run pattern
  • MkDocs configuration with Material theme and advanced search

View Full Release Notes →

Previous Releases

Version 0.2.3 (January 29, 2026)

Architecture Evolution: Major refactoring to split cuvis-ai into core framework (cuvis-ai-core) and algorithm catalog (cuvis-ai).

Key Changes:

  • Plugin system with dynamic loading from Git repositories and local filesystem
  • 426 tests migrated to cuvis-ai-core with independent CI/CD
  • Session-scoped plugin isolation for gRPC services
  • Import pattern changes to cuvis_ai_core.*
  • New gRPC RPCs for plugin management

View Full Release Notes →

Version 0.2.2 (January 15, 2026)

Restoration Utilities: Simplified pipeline and trainrun restoration with auto-detection.

Key Changes:

  • New CLI commands: restore-pipeline, restore-trainrun
  • Auto-detection of statistical vs gradient training workflows
  • Consolidated restoration utilities in single module
  • Standardized Python API surface

View Full Release Notes →

Version 0.2.1 (January 8, 2026)

Configuration System: Pydantic v2 models and server-side Hydra composition.

Key Changes:

  • Pydantic v2 config models as single source of truth
  • Session-scoped configuration management
  • Terminology change: Experiment → TrainRun
  • Explicit 4-step workflow for gRPC API
  • Standardized config transport with config_bytes

View Full Release Notes →

Version 0.2.0 (December 20, 2025)

YAML-Driven Pipelines: Introduced configuration-based pipeline building.

Key Changes:

  • YAML-driven pipeline configuration with OmegaConf
  • Hybrid NodeRegistry for built-ins and custom nodes
  • End-to-end serialization (YAML + .pt weights)
  • Version compatibility guards
  • Simplified node state management

View Full Release Notes →

Version 0.1.5 (December 1, 2025)

gRPC Services: Introduced remote API for pipeline execution.

Key Changes:

  • gRPC service stack with proto definitions
  • Buf Schema Registry integration
  • Session management and PipelineBuilder
  • Two-phase training support
  • Pipeline introspection RPCs

View Full Release Notes →

Version 0.1.3 (November 6, 2025)

Port-Based I/O: Major refactor introducing typed input/output system.

Key Changes:

  • Port-based typed I/O with PortSpec
  • Graph connection API with auto-validation
  • Multi-input/output support
  • PyTorch Lightning integration
  • Executor refactored for port-based routing

View Full Release Notes →

Versioning

cuvis-ai follows Semantic Versioning (MAJOR.MINOR.PATCH):

  • MAJOR: Incompatible API changes
  • MINOR: New functionality (backwards-compatible)
  • PATCH: Bug fixes (backwards-compatible)

Release Channels

  • Stable: Use tagged releases for production deployments
  • Latest: Track the main branch for latest stable code
  • Development: Track feature branches for experimental features

See Also