• noarch v0.2.1

pip install

conda install

anaconda-assistant

The Anaconda Assistant Python client

Installation

conda install -c anaconda-cloud anaconda-assistant-sdk

How to authenticate

To use the Python client or CLI you can use anaconda login CLI, Anaconda Navigator, or

from anaconda_cloud_auth import login
login()

to launch a browser to login and save your API token to disk. For cases where you cannot utilize a browser to login you can grab your API and set the ANACONDA_CLOUD_API_KEY=<api-key> env var.

The Python clients and integrations provide api_key as a keyword argument.

Terms of use and data collection

In order to use the Anaconda Assistant SDK and derived integrations the user must first agree to

• Our terms of service
• Our privacy policy
• Assert that they are more than 13 years old
• Opt-in or Opt-out of Data Collection

On Data Collection: If you opt-in you will enjoy personalized recommendations and contribute to smarter features.

We prioritize your privacy:

• Your data is never sold
• Always secured
• This setting only affects the data Anaconda stores
• It does not affect the data that is sent to Open AI

To agree to the terms of service and configure data collection edit the ~/.anaconda/config.toml file

[plugin.assistant]
accepted_terms = true
data_collection = true

You may set data_collection = false if you chose to opt-out.

If you set accepted_terms = false the Anaconda Assistant SDK and derived integrations will not function.

If either or both of these values are unset in the ~/.anaconda/config.toml file, an exception will be raised.

Chat session

The ChatSession provides a multi-turn chat interface that saves input and output messages. The response can be streamed or provided in one chunk.

from anaconda_assistant import ChatSession

chat = ChatSession()

text = chat.completions("what are the the first 5 fibonacci numbers?", stream=False)
print(text)

text = chat.completions("make that the first 10 numbers", stream=True)
for chunk in text:
    print(chunk, end="")

Chat client

The ChatClient provides a low-level completions function that accepts a list of messages in the same format as OpenAI. The completions() method returns a ChatResponse object that allows streaming of the response similar to requests Response with .iter_content(), .iter_lines(), and .message, which returns the whole message as a string. Once the who message has been access or consumed through an iterable it is retained in the .message object. Additionally, the response object stores .tokens_used and .tokens_limit integers.

from anaconda_assistant import ChatClient

client = ChatClient()

messages = [
    {"role": "user", "content": "What is pi?"}
]

response = client.completions(messages=messages)

for chunk in response.iter_content():
    print(chunk, end="")

You can only consume the message with .iter_content() once, but the result is captured to the .message attribute while streaming.

Daily quotas

Each Anaconda.cloud subscription plan enforces a limit on the number of requests (calls to .completions()). The limits are documented on the Plans and Pricing page. Once the limit is reached the .completions() function will throw a DailyQuotaExceeded exception.

Users can upgrade their plans by visiting https://anaconda.cloud/profile/subscriptions.

Integrations

A number of 3rd party integrations are provided. In each case you will need to have optional packages installed.

LLM CLI

The LLM CLI can be use to send and receive messages with the Anaconda Assistant.

Required packages: llm

To direct your messages to Anaconda Assistant use the model name anaconda-assistant

> llm -m anaconda-assistant 'what is pi?'

LangChain

A LangChain integration is provided that supports message streaming and non-streaming responses.

Required packages: langchain-core >=0.3

from anaconda_assistant.langchain import AnacondaAssistant
from langchain.prompts import ChatPromptTemplate

model = AnacondaAssistant()
prompt = ChatPromptTemplate.from_template("tell me a joke about {topic}")
chain = prompt | model

message = chain.invoke({'topic': 'python'})
print(message.content)

ELL

You can use Anaconda Assistant as a model in the ell prompt engineering framework.

Required packages: ell-ai

import ell
import anaconda_assistant.integrations.ell

ell.init(verbose=True)

@ell.simple(model="anaconda-assistant")
def who():
    return "Who are you?"

who()

PandasAI

To use Anaconda Assistant with PandasAI configure the SmartDataFrame using the AnacondaAssistant plugin

Required packages: pandasai

from anaconda_assistant.pandasai import AnacondaAssistant
from pandasai import SmartDataframe

ai = AnacondaAssistant()
sdf = SmartDataframe(df, config={'llm': ai})
sdf.chat('what is the average of this column where some condition is true?')

Panel

You can integrate the Anaconda Assistant in your Panel application using the chat features including the name and avatar for the signed in user.

Required packages: panel

import panel as pn

from anaconda_assistant import ChatSession
from anaconda_assistant.integrations.panel import AnacondaAssistantCallbackHandler

assistant = ChatSession()
callback = AnacondaAssistantCallbackHandler(assistant)

chat = pn.chat.ChatInterface(
    callback=callback,
    user=assistant.auth_client.name,
    avatar=assistant.auth_client.avatar,
)

Setup for development

Ensure you have conda installed. Then run:

make setup

Run the unit tests

make test

Run the unit tests across isolated environments with tox

make tox