v0.3.1

Exposes the OpenAI instance being used through OpenAIChatStream.client()

v0.3.0

This releases updates the implementation of OpenAI, LiteLLM and GPT4All for their latest versions, there are no breaking changes for LangStream since LangStream API wraps them entirely, except for the removal of the stream parameter on GPT4All, since it's no longer necessary to disable it to get replicate and others to work properly.

v0.2.0

This release makes LLM implementation as optional dependencies, fixing #11, it might be a breaking change for you

Now, if you want to use langstream with say, OpenAI, you need to install the openai package separately, if you want to use with gpt4all or litellm, then you also need to install their packages separately, and so on

v0.1.8

• Add support to LiteLLM, bringing Azure, Anthropic, HuggingFace and many others to LangStream #10 (docs)

v0.1.7

BREAKING: LiteChain was renamed to LangStream! For more details check out #4

To migrate it should be easy enough, change the dependency and imports from litechain to langstream, and replace all matches of chain to stream and matches of Chain to Stream in your project

v0.1.6

• Add .pipe() operator 26033e8 (docs)
• Add .filter() operator 33b1800 (docs)

v0.1.5

• Put a short timeout and retry mechanism on OpenAI chains to make them more reliable, since API hangs or failures are not very unusual (see 738b707)

v0.1.4

• Add on_error method on chains for error handling (See #3)
• Removed unused generic type from ChainOutput (See #1)

v0.1.3

BREAKING: This release REMOVES the "magical" py2gpt function conversion on OpenAI function calls, it accepts only raw OpenAI function schemas now, more details on #2

v0.1.2

• Add name key to OpenAIChatMessage and OpenAIChatDelta, as it is necessary for function calls and storing it back on memory - thanks rbdixon on Discord!