- Add new fusion strategies for the hybrid vector store: RRF and DBSF (#413)
- move sources from ragbits-document-search to ragbits-core (#496)
- adding connection check to Azure get_blob_service (#502)
- Make the score in VectorStoreResult consistent (always bigger is better)
- Add router option to LiteLLMEmbedder (#440)
- Make LLM / Embedder APIs consistent (#463)
- New methods in Prompt class for appending conversation history (#480)
- Fix: make unflatten_dict symmetric to flatten_dict (#461)
- Cost and capabilities config for custom litellm models (#481)
- Allow Prompt class to accept the asynchronous response_parser. Change the signature of parse_response method.
- Fix from_config for LiteLLM class (#441)
- Fix Qdrant vector store serialization (#419)
- Add HybridSearchVectorStore which can aggregate results from multiple VectorStores (#412)
- Better handling of cases when text and image embeddings are mixed in VectorStore
- BREAKING CHANGE: Vector Stores are now responsible for creating embeddings (#288)
- Qdrant vector store can now be serialized during Ray processing (#394)
- Improve cli trace handler
- Add traceable to some method
- Add support for images in few shot prompts (#155)
- Add instruction on how to use local servers for LLMs (#395).
- Introduce intermediate image elements (#139)
- Correct typos in doc strings (#398)
- Enable GPU support and (un)pickling for fastembed embedders (#409).
- Add support to fastembed dense & sparse embeddings.
- Rename "default configuration" to "preferred configuration" (#361).
- Allow to pass str or dict to LLM.generate() (#286)
- Fix: changed variable type from Filter to WhereQuery in the Qdrant vector store in list method.
- Rename all embedders to have
Embedder
in their name (instead ofEmbeddings
).
- Add support for pgvector as VectorStore (#267).
- Add nice-looking CLI logging for audit module (#273).
- Add support for returning metadata from LLMs (#274).
- Fix: limiting in qdrant vector store (#282).
- Refactor: remove LLM client abstraction and lift it up to LLM (#270).
- Add option to pass LiteLLM router to LLM instances (#262).
- Add commands to browse vector stores (#244).
- Implement generic Options class (#248).
- Fix LiteLLM crash in python 3.13 (#245).
- Refactor: added standardized way to create ragbits objects from config (#233).
- Default LLM factory when configuration is not provided (#209).
- Add remove operation to VectorStore (#210).
- Install litellm package by default (#236).
- Add support for Qdrant VectorStore (#163).
- Add streaming interface to LLMs (#188).
- Better images support in Prompt abstractions (#201).
- Observability toolset, with initial support to export traces to OpenTelemetry (#168)
- CLI commands to render / exec prompts (#146)
- Support of images in Prompt abstractions (#149)
- Support for different MetadataStores in VectorStore (#144)
- Now LLMs can be configured separately for vision, text and structured. (#153)
- refactor: Add dynamic loading for modules that depend on optional dependencies (#148).
- refactor: Refactor vector store public API. (#151)
- Project README.md (#103).
- Listing entries API for VectorStores (#138).
- Overrides for prompt discovery configurable in
pyproject.toml
file (#101). - Default LLM factory configurable in
pyproject.toml
file (#101).
- Fixed bug in chromadb while returning multiple records (#117).
- Fixed bug in prompt rendering for some pydantic models (#137).
- Initial release of the package.
- Introduce core components: Prompts, LLMs, Embedder and VectorStores.
Prompt
class integration with promptfoo.- LiteLLM integration.
- ChromaDB integration.
- Prompts lab.
- Prompts autodiscovery.