Skip to content

v0.5.0 - Resilient Batch Indexing and SDK/CLI Parity

Choose a tag to compare

@github-actions github-actions released this 12 Feb 08:24
· 95 commits to master since this release
654b696

Added

  • Structured batch indexing result models for SDK consumers:
    • BatchIndexResult (inputs, discovered/indexed/skipped sources, failures)
    • BatchIndexFailure (source_url, stage, error_message)
  • New AudioRAGPipeline.index_many(inputs, force=False, raise_on_error=True)
    contract that returns structured batch outcomes and supports tolerant mode
    (raise_on_error=False) for partial-failure reporting.

Changed

  • Core indexing boundary now performs source discovery for SDK calls too:
    • index(url) routes through discovery-backed batch indexing
    • Playlist/channel URLs and local directories are processed per source with
      independent state/resume semantics.
  • Batch indexing error handling is now resilient across exception types:
    non-PipelineError failures are normalized and captured in batch results,
    while strict mode still raises with clear context.
  • CLI audiorag index now uses the unified pipeline batch path and reports
    aggregate batch outcomes (indexed/skipped/failed) with per-source failures.

Documentation

  • Updated README and docs (quickstart, api-reference) to reflect:
    • SDK/CLI behavior parity for discovery-backed indexing
    • raise_on_error behavior and structured batch results
    • Updated batch indexing examples and output expectations