Where Does LangChain Define Its Chat Templates?

LangChain doesn’t store those special tokens or templates as separate plaintext files on disk or in dedicated config files. Instead, the chat templates (and their placeholders, like the ones representing a user’s message) are defined as Python objects within the LangChain library. When you create a chain or agent, LangChain constructs the necessary prompt templates dynamically using classes such as ChatPromptTemplate, HumanMessagePromptTemplate, and others. These classes encapsulate the formatting rules and token handling for the underlying chat models.

In other words, the tokens you mention (like "<|user|>") aren’t found by searching through the repository because they’re either abstractions or they come from the chat model providers' internal implementations, not from a static plaintext file in LangChain. The LangChain source code defines their behavior through code, which is why you can inspect how they work by examining the relevant modules (for example, in chat.py or related prompt modules), but you won’t find a file that contains a literal listing of those tokens.