Using the Chat API

The Chat API endpoint is used to generate text with Cohere LLMs. This endpoint facilitates a conversational interface, allowing users to send messages to the model and receive text responses.

1import cohere
2co = cohere.Client(api_key="<YOUR API KEY>")
3
4response = co.chat(
5  model="command-r-plus",
6  message="Write a title for a blog post about API design. Only output the title text."
7)
8
9print(response.text) # "The Art of API Design: Crafting Elegant and Powerful Interfaces"

Response Structure

Below is a sample response from the Chat API

JSON
1{
2    "text": "The Art of API Design: Crafting Elegant and Powerful Interfaces",
3    "generation_id": "dd78b9fe-988b-4c18-9419-8fbdf9968948",
4    "chat_history": [
5        {
6            "role": "USER",
7            "message": "Write a title for a blog post about API design. Only output the title text."
8        },
9        {
10            "role": "CHATBOT",
11            "message": "The Art of API Design: Crafting Elegant and Powerful Interfaces"
12        }
13    ],
14    "finish_reason": "COMPLETE",
15    "meta": {
16        "api_version": {
17            "version": "1"
18        },
19        "billed_units": {
20            "input_tokens": 17,
21            "output_tokens": 12
22        },
23        "tokens": {
24            "input_tokens": 83,
25            "output_tokens": 12
26        }
27    }
28}

Every response contains the following fields:

  • text the generated message from the model.
  • generation_id the ID corresponding to this response. Can be used together with the Feedback API endpoint to promote great responses and flag bad ones.
  • chat_history the conversation presented in a chat log format
  • finish_reason can be one of the following:
    • COMPLETE the model successfully finished generating the message
    • MAX_TOKENS the model’s context limit was reached before the generation could be completed
  • meta contains information with token counts, billing etc.

Multi-turn conversations

The user message in the Chat request can be sent together with a chat_history to provide the model with conversational context:

PYTHON
1import cohere
2co = cohere.Client(api_key="<YOUR API KEY>")
3
4message = "Can you tell me about LLMs?"
5
6response = co.chat(
7  model="command-r-plus",
8	chat_history=[
9    {"role": "USER", "text": "Hey, my name is Michael!"},
10    {"role": "CHATBOT", "text": "Hey Michael! How can I help you today?"},
11  ],
12  message=message
13)
14
15print(response.text) # "Sure thing Michael, LLMs are ..."

Instead of manually building the chat_history, we can grab it from the response of the previous turn.

PYTHON
1chat_history = []
2max_turns = 10
3
4for _ in range(max_turns):
5	# get user input
6	message = input("Send the model a message: ")
7	
8	# generate a response with the current chat history
9	response = co.chat(
10    model="command-r-plus",
11		message=message,
12		chat_history=chat_history
13	)
14  
15  # print the model's response on this turn  
16	print(response.text)
17  
18	# set the chat history for next turn
19	chat_history = response.chat_history

Using conversation_id to Save Chat History

Providing the model with the conversation history is one way to have a multi-turn conversation with the model. Cohere has developed another option for users who do not wish to save the conversation history, and it works through a user-defined conversation_id.

PYTHON
1import cohere
2co = cohere.Client("<YOUR API KEY>")
3
4response = co.chat(
5	model="command-r-plus",
6	message="The secret word is 'fish', remember that.",
7	conversation_id='user_defined_id_1',
8)
9
10answer = response.text

Then, if you wanted to continue the conversation, you could do so like this (keeping the id consistent):

PYTHON
1response2 = co.chat(
2	model="command-r-plus",
3	message="What is the secret word?",
4	conversation_id='user_defined_id_1'
5)
6
7print(response2.text) # "The secret word is 'fish'"

Note that the conversation_id should not be used in conjunction with the chat_history. They are mutually exclusive.