feat(prompts): chat message placeholders #1222
Conversation
![greptile-apps[bot]](https://avatars.githubusercontent.com/in/867647?s=60&v=4){kind=small}
There was a problem hiding this comment.
15 files reviewed, 1 comment
Edit PR Review Bot Settings | Greptile
langfuse/model.py
Outdated
| for chat_message in messages_with_placeholders_replaced | ||
| if "role" in chat_message and "content" in chat_message | ||
| ] | ||
|
|
||
| def get_langchain_prompt(self, **kwargs): |
There was a problem hiding this comment.
we need to cover langchain prompts as well where users are not compiling prompts with langfuse but with langchain. In this case, they might want to only fill in placeholders and keep variables untouched and then call get_langchain_prompt.
We can either add a separate method that is only handling replacing placeholders, or we rely on compileWithPlaceholders(variables={}, placeholders=...) but we need a test here to make sure this works as expected
![ellipsis-dev[bot]](https://avatars.githubusercontent.com/in/64358?s=60&v=4){kind=small}
| ) -> None: | ||
| """Backward-compatible setter for raw prompt structure.""" | ||
| for p in prompt: | ||
| if hasattr(p, "type") and hasattr(p, "name") and p.type == "placeholder": |
There was a problem hiding this comment.
Using hasattr(p, ...) in the prompt setter may fail for plain dict inputs which lack attribute access. Consider checking if p is a dict (e.g. using 'if isinstance(p, dict) and "type" in p') to support both dicts and objects.
| err_placeholder_not_list = f"Placeholder '{chat_message['name']}' must contain a list of chat messages, got {type(placeholder_messages)}" | ||
| raise ValueError(err_placeholder_not_list) | ||
| else: | ||
| compiled_messages.append( |
There was a problem hiding this comment.
Unresolved placeholders are logged as warnings and a placeholder dict (with only 'type' and 'name') is appended. This may later cause KeyError in methods expecting 'role' and 'content'. Consider either raising an error or handling them explicitly.
Important
Adds support for chat message placeholders in langfuse-python SDK, enabling dynamic templating in prompts.
ChatMessageWithPlaceholderstype inchat_message_with_placeholders.pyto handle both regular and placeholder messages.create_prompt_request.pyandchat_prompt.pyto useChatMessageWithPlaceholdersfor prompt creation.PlaceholderMessagemodel inplaceholder_message.pyfor template variables.create_promptmethod inclient.pyto acceptChatMessageWithPlaceholders.test_prompt.pyandtest_prompt_compilation.pyfor placeholder scenarios and edge cases.This description was created by
for e295821. You can customize this summary. It will automatically update as commits are pushed.
Greptile Summary
Disclaimer: Experimental PR review
Implements chat message placeholder functionality in the langfuse-python SDK, enabling dynamic templating in chat prompts. This feature is critical for scenarios like injecting conversation history or examples into prompts.
ChatMessageWithPlaceholdersunion type inlangfuse/api/resources/prompts/types/chat_message_with_placeholders.pysupporting both regular chat messages and placeholder messagesChatMessageWithPlaceholdersinstead ofChatMessageinlangfuse/api/resources/prompts/types/create_prompt_request.pyPlaceholderMessagePydantic model inlangfuse/api/resources/prompts/types/placeholder_message.pyfor template variablestests/test_prompt.pyfor placeholder scenarios and edge cases