0.0.14

⚠️ Breaking Changes ⚠️

• Removed memory.ConversationBufferWindowMemory and memory.ConversationBufferMemory. They have been replaced with memory.ChatMessageHistory.

In order to keep up with the changes in upstream LangChain, we converted our Memory classes to a single ChatMessageHistory that may be used with the upstream conversational memory classes. This is a bit unfortunate, but should allow us to continue to stay in-step with the latest LangChain developments.

We apologize for the impact.

What's Changed

• loaders: support allowed failures in bulk operations by @douglas-reid in #23
• feat: add support for direct File import to VectorStore by @douglas-reid in #25
• feat: add file loader for sphinx-based sites by @douglas-reid in #26
• deps: update to match latest LC memory refactor by @douglas-reid in #28
• feat: allow sphinx loader to sanitize and ignore by @douglas-reid in #27

Full Changelog: 0.0.13...0.0.14

v0.0.13 - Hotfix for VectorStore deployments

What's Changed

• Hotfix steamship vector store by @EniasCailliau in #22

Full Changelog: 0.0.12...0.0.13

v0.0.12

What's Changed

• feat: add logging callback by @douglas-reid in #18
• docs: add information on logging callback by @steamship-developers in #20

Full Changelog: 0.0.11...0.0.12

Add logging callback with doc fixes

Merge pull request #16 from steamship-core/doc-fix

docs: fix document_loaders index

v0.0.11 - VectorStore and Loaders

What's Changed

• Update embedding model to text-embedding-ada-002 by @EniasCailliau in #8
• Enable in place replacement of langchain.OpenAI with steamship_langchain.OpenAI by @EniasCailliau in #4
• First version of a SteamshipVectorStore by @EniasCailliau in #5
• feat: add initial set of file_loaders by @douglas-reid in #10
• feat: readthedocs support by @douglas-reid in #14

New Contributors

• @EniasCailliau made their first contribution in #8

Full Changelog: 0.0.10...0.0.11

Full OpenAI support

This release updates the LLM support to provide a full drop-in replacement for the OpenAI LLM in LangChain. This allows users to use the Steamship backend for LLM calls with only a package name change. Thanks @EniasCailliau for the contribution.

Better support for plugin retry logic in client

Includes more forgiving wait() time in Task handling around generation. This will work better with the updated retry / backoff behavior in the LLM plugin.