### Install LangSmith Python SDK

Source: https://github.com/langchain-ai/chat-langchain/blob/master/LANGSMITH.md

Before running evaluations or interacting with LangSmith programmatically, you need to install the LangSmith Python SDK. Use pip to add the 'langsmith' package to your development environment.

```shell
pip install langsmith
```

--------------------------------

### Instantiate Weaviate Vector Store in Python

Source: https://github.com/langchain-ai/chat-langchain/blob/master/MODIFY.md

This code snippet demonstrates how to instantiate a Weaviate client and vector store within the `ingest_docs` function in `backend/ingest.py` for Chat LangChain. It shows how to configure the client with a URL and API key, and then initialize the `Weaviate` vector store with the client, index name, text key, embedding, and attributes. This setup is crucial for storing document embeddings for ingestion and retrieval operations, and similar steps apply to `backend/chain.py`.

```python
client = weaviate.Client(
    url=WEAVIATE_URL,
    auth_client_secret=weaviate.AuthApiKey(api_key=WEAVIATE_API_KEY),
)
vectorstore = Weaviate(
    client=client,
    index_name=WEAVIATE_DOCS_INDEX_NAME,
    text_key="text",
    embedding=embedding,
    by_text=False,
    attributes=["source", "title"],
)
```

--------------------------------

### Build Chat LangChain Frontend Locally

Source: https://github.com/langchain-ai/chat-langchain/blob/master/DEPLOYMENT.md

Commands to navigate into the frontend directory and build the application locally. This step ensures the frontend is ready before deployment to Vercel.

```shell
cd frontend
yarn
yarn build
```

--------------------------------

### GitHub Actions: Configure Environment Secrets for Indexing

Source: https://github.com/langchain-ai/chat-langchain/blob/master/DEPLOYMENT.md

Environment variables required for the 'Indexing' environment in GitHub Actions. These secrets are crucial for recurring data ingestion tasks and should match those in Vercel.

```APIDOC
OPENAI_API_KEY=
RECORD_MANAGER_DB_URL=
WEAVIATE_API_KEY=
WEAVIATE_INDEX_NAME=langchain
WEAVIATE_URL=
```

--------------------------------

### Vercel: Connect Frontend to LangGraph Cloud Backend API

Source: https://github.com/langchain-ai/chat-langchain/blob/master/DEPLOYMENT.md

Environment variables to be added in Vercel for connecting the frontend to the LangGraph Cloud backend API. This includes the base API URL, the public-facing API URL, and the LangSmith API key for monitoring.

```APIDOC
API_BASE_URL: Matches your LangGraph Cloud deployment API URL
NEXT_PUBLIC_API_URL: API URL that LangGraph Cloud deployment is proxied to, e.g. "https://chat.langchain.com/api"
LANGCHAIN_API_KEY: LangSmith API key
```

--------------------------------

### Python Prompt Template for Rephrasing Follow-up Questions

Source: https://github.com/langchain-ai/chat-langchain/blob/master/CONCEPTS.md

This Python string defines a prompt template used to rephrase a follow-up question into a standalone question. It takes the chat history and the current follow-up question as input, enabling the language model to generate a semantically clearer query for better vector store searches.

```python
REPHRASE_TEMPLATE = """Given the following conversation and a follow up question, rephrase the follow up question to be a standalone question.\n\nChat History:\n{chat_history}\nFollow Up Input: {question}\nStandalone Question:"""
```

--------------------------------

### Configure LangSmith Tracing Environment Variables

Source: https://github.com/langchain-ai/chat-langchain/blob/master/LANGSMITH.md

To enable LangSmith tracing for your LLM application, set the LANGCHAIN_TRACING_V2 environment variable to true and provide your LangChain API key as LANGCHAIN_API_KEY. These variables are the simplest way to activate observability.

```shell
export LANGCHAIN_TRACING_V2=true
export LANGCHAIN_API_KEY=...
```

--------------------------------

### LangChain Indexing API for RAG Applications

Source: https://github.com/langchain-ai/chat-langchain/blob/master/CONCEPTS.md

Outlines the importance of indexing in production RAG applications to prevent duplicate documents and improve performance. Describes the two main steps: Ingestion and Hashing, and how the LangChain indexing API facilitates robust document management.

```APIDOC
LangChain Indexing API:
  Purpose: Robust document indexing for production RAG applications.
  Benefits:
    - Prevents Duplicate Results: Ensures unique documents in vector store.
    - Improves Performance: Only new documents require embedding generation.
  Process Steps:
    1. Ingestion: Pulls documents to be added.
    2. Hashing: Creates unique hashes for each document (with metadata) to identify new documents.
    3. Insertion: Inserts new, hashed documents into the vector store (after Record Manager check).
```

--------------------------------

### Instantiate MongoDocumentManager for Record Management in Python

Source: https://github.com/langchain-ai/chat-langchain/blob/master/MODIFY.md

This code snippet illustrates how to replace the default `SQLRecordManager` with a `MongoDocumentManager` for document ingestion in Chat LangChain. It demonstrates importing `MongoDocumentManager` and then instantiating it with a namespace, MongoDB URL, database name, and collection name. The `create_schema()` method is called to initialize the necessary schema for record management.

```python
from langchain_community.indexes import MongoDocumentManager

record_manager = MongoDocumentManager(
    namespace="kittens",
    mongodb_url="mongodb://langchain:langchain@localhost:6022/",
    db_name="test_db",
    collection_name="test_collection",
)
record_manager.create_schema()
```

--------------------------------

### Understanding Vector Stores in Chat LangChain

Source: https://github.com/langchain-ai/chat-langchain/blob/master/CONCEPTS.md

Explains the fundamental role of vector stores as specialized databases for high-dimensional vectors, generated by text embedding models. Details the process of vector generation, similarity searches, and context retrieval to enhance language model capabilities.

```APIDOC
VectorStore:
  Description: Specialized database for high-dimensional vectors from text embedding models.
  Role in Chat LangChain:
    - Vector Generation: Converts textual content into unique vectors (fingerprints).
    - Similarity Searches: Finds documents relevant to a user query by comparing vector distances.
    - Context Retrieval: Provides retrieved documents to the language model for informed responses.
```

--------------------------------

### Add New LLM Alternatives and Update Response Synthesizer

Source: https://github.com/langchain-ai/chat-langchain/blob/master/MODIFY.md

This snippet shows how to extend the existing `configurable_alternatives` method of the `llm` variable by adding a new LLM class declaration. It also demonstrates the necessary step of updating the `response_synthesizer` variable to include the newly added LLM alternative, ensuring it's properly integrated into the response generation pipeline.

```Python
.configurable_alternatives(
    local_ollama=ChatCohere(
        model="llama2",
        temperature=0,
    ),
)
```

```Python
response_synthesizer = (
    default_response_synthesizer.configurable_alternatives(
        ConfigurableField("llm"),
        default_key="openai_gpt_3_5_turbo",
        anthropic_claude_3_haiku=default_response_synthesizer,
        ...
        local_ollama=default_response_synthesizer,
    )
    | StrOutputParser()
).with_config(run_name="GenerateResponse")
```

--------------------------------

### LangChain Record Manager API Overview

Source: https://github.com/langchain-ai/chat-langchain/blob/master/CONCEPTS.md

Provides an interface for managing records of upserted documents before ingestion into a vector store. Details key concepts like Namespace, Keys, Group IDs, and Timestamps, enabling efficient tracking, insertion, update, deletion, and querying of records.

```APIDOC
LangChain Record Manager API:
  Interface: Manages records of upserted documents before vector store ingestion.
  Capabilities: Efficiently insert, update, delete, and query records.
  Core Concepts:
    - Namespace: Logical separation of records.
    - Keys: Unique identifier for each record.
    - Group IDs: Optional grouping for batch operations.
    - Timestamps: Tracks last update time (updated_at) for time-range queries.
  Function: Tracks which documents need to be added/updated in the vector store, preventing duplication.
```

--------------------------------

### LangChain Smith Evaluation run_on_dataset Function

Source: https://github.com/langchain-ai/chat-langchain/blob/master/LANGSMITH.md

The `run_on_dataset` function from `langchain.smith.evaluation.runner_utils` is used to evaluate your LLM application against a predefined dataset. It allows you to configure custom evaluation criteria, such as semantic similarity or using an LLM as a judge, and then view the results in the LangSmith dashboard.

```APIDOC
Function: langchain.smith.evaluation.runner_utils.run_on_dataset
  Purpose: Evaluates an LLM application against a given dataset using specified criteria.
  Parameters:
    dataset: (Type not specified) The dataset to be used for evaluation.
    evaluation_criteria: (Type not specified) Custom criteria for evaluation (e.g., semantic similarity, LLM as a judge).
  Returns: (Type not specified) Evaluation results, viewable in the LangSmith dashboard.
```

--------------------------------

### Customize LangChain Runnable Run Name with .with_config

Source: https://github.com/langchain-ai/chat-langchain/blob/master/LANGSMITH.md

For more detailed tracing in LangSmith, you can add a custom run name to any LangChain Runnable using the .with_config method. This helps in organizing and identifying specific operations within your traces, especially for extending default tracing behavior.

```python
.with_config(
    run_name="CondenseQuestion",
)
```

--------------------------------

### Replace All LLM Providers with a Single LangChain LLM

Source: https://github.com/langchain-ai/chat-langchain/blob/master/MODIFY.md

This snippet demonstrates how to replace the entire `llm` variable declaration in `./backend/chain.py` with a single LangChain LLM class. It simplifies the LLM configuration by removing multiple configurable alternatives and using a single chosen LLM, while maintaining the `llm` variable name to avoid breaking other parts of the endpoint.

```Python
llm = ChatOpenAI(
    model="gpt-4o-mini-2024-07-18",
    streaming=True,
    temperature=0,
).configurable_alternatives(
    # This gives this field an id
    # When configuring the end runnable, we can then use this id to configure this field
    ConfigurableField(id="llm"),
    default_key="openai_gpt_3_5_turbo",
    anthropic_claude_3_haiku=ChatAnthropic(
        model="claude-3-haiku-20240307",
        max_tokens=16384,
        temperature=0,
        anthropic_api_key=os.environ.get("ANTHROPIC_API_KEY", "not_provided"),
    ),
    ...
)
```

```Python
llm = ChatYourLLM(
    model="model-name",
    streaming=True,
    temperature=0,
).configurable_alternatives(
    # This gives this field an id
    ConfigurableField(id="llm")
)
```

--------------------------------

### Change Embeddings Model in Chat LangChain

Source: https://github.com/langchain-ai/chat-langchain/blob/master/MODIFY.md

This snippet illustrates how to modify the `get_embeddings_model` function in `./backend/ingest.py` to use a different embeddings model. It demonstrates replacing the default `OpenAIEmbeddings` with `MistralAIEmbeddings`, requiring an import and API key configuration, thereby allowing customization of the vector store's embedding process.

```Python
def get_embeddings_model() -> Embeddings:
    return OpenAIEmbeddings(model="text-embedding-3-small", chunk_size=200)
```

```Python
from langchain_mistralai import MistralAIEmbeddings

def get_embeddings_model() -> Embeddings:
    return MistralAIEmbeddings(mistral_api_key="your-api-key")
```

=== COMPLETE CONTENT === This response contains all available snippets from this library. No additional content exists. Do not make further requests.