Cohere LangChain Integration

0.5.0 · active · verified Sun Apr 12

langchain-cohere is an integration package that connects Cohere's powerful language AI models (Chat, Embeddings, Rerank) with the LangChain framework. It simplifies the process of building LLM-powered applications using Cohere's capabilities. The library is actively developed, with frequent releases often occurring on a weekly or bi-weekly cadence, and the current version is 0.5.0.

Warnings

breaking The `create_csv_agent` abstraction was deprecated in v0.3.5 and subsequently removed in v0.4.1. Users relying on this function will need to refactor their code.
Fix: Migrate to using `create_csv_agent` from `langchain_experimental.agents.agent_toolkits.csv.base` and manually construct the agent workflow.
breaking Cohere integrations were moved from `langchain_community` (and other `langchain` paths) to the dedicated `langchain-cohere` package. Direct imports from `langchain_community.chat_models.ChatCohere`, `langchain.embeddings.CohereEmbeddings`, or `langchain.retrievers.document_compressors.CohereRerank` are deprecated and will break in newer `langchain` versions.
Fix: Update import statements to use `from langchain_cohere import ...` for all Cohere components. For example, `from langchain_community.chat_models import ChatCohere` should become `from langchain_cohere import ChatCohere`.
breaking The Cohere Chat and Rerank APIs migrated to v2 starting from v0.4.1. This introduced significant changes to message structure (using a single `messages` parameter with roles), chat history handling, tool definitions (JSON schema), and requires the `model` parameter to be explicitly specified for all calls.
Fix: Review Cohere API v2 migration guides for `Chat` and `Rerank` to adapt request formats, message structures, and ensure the `model` parameter is always provided. `CohereRerank` now mandates the `model` parameter.
deprecated The `create_cohere_tools_agent` and general `connectors` functionality were deprecated in v0.3.3.
Fix: Migrate to `create_cohere_react_agent` found at `langchain_cohere.react_multi_hop.agent`. Re-evaluate usage of `connectors` as their functionality may have changed or been replaced by alternative methods within the Cohere API or LangChain.
gotcha The `langchain-cohere` package now manages the compatible `cohere` Python SDK version (v5+). Using an older or incompatible `cohere` SDK version installed separately may lead to unexpected behavior or errors.
Fix: Ensure that your installed `cohere` package is version 5 or higher. `pip install --upgrade cohere` may be necessary. It's recommended to let `langchain-cohere` manage this dependency.

Install

pip install langchain-cohere Install package

Imports

ChatCohere
```
from langchain_cohere import ChatCohere
```
Cohere chat model integrations were moved from `langchain_community` to `langchain_cohere`.
CohereEmbeddings
```
from langchain_cohere import CohereEmbeddings
```
Embeddings integrations were moved from `langchain.embeddings` (and `langchain_community`) to `langchain_cohere`.
CohereRerank
```
from langchain_cohere import CohereRerank
```
Rerank integrations were moved from `langchain.retrievers.document_compressors` (and `langchain_community`) to `langchain_cohere`.
create_cohere_react_agent
```
from langchain_cohere.react_multi_hop.agent import create_cohere_react_agent
```
`create_cohere_tools_agent` was deprecated in v0.3.3 and has been replaced by `create_cohere_react_agent` with a different import path.

Quickstart

This quickstart demonstrates how to initialize the `ChatCohere` model and perform a simple chat completion. It uses the `COHERE_API_KEY` environment variable for authentication and `command-r-plus` as the model, which can be changed to other supported Cohere models.

import os
from langchain_cohere import ChatCohere
from langchain_core.messages import HumanMessage

# Ensure your COHERE_API_KEY is set as an environment variable
# or pass it directly to the ChatCohere constructor.
# For example: cohere_api_key="your_api_key"

llm = ChatCohere(
    model="command-r-plus", # or another supported Cohere chat model
    temperature=0, 
    cohere_api_key=os.environ.get('COHERE_API_KEY', '')
)

messages = [HumanMessage(content="What is the capital of Canada?")]

try:
    response = llm.invoke(messages)
    print(response.content)
except Exception as e:
    print(f"An error occurred: {e}")
    print("Please ensure COHERE_API_KEY is set and the model name is correct.")

view raw JSON →