Skip to main content

Unofficial Anthropic Claude API for Python3.

Project description

Unofficial Claude API For Python

The UNOFFICIAL free API for Anthropic's Claude LLM.

Background

Claude is Anthropic's LLM app (similar to ChatGPT). This library allows you to use the API (for free) and interact with it in your python projects.

Implemented actions:

The Unofficial Claude API is under active development. The following endpoints are usable in some capacity:

  • Getting organizations you're in
  • Getting conversations you're in
  • Starting a conversation
  • Sending a message and receiving a response
  • Delete a conversation
  • Create an attachment from a file
  • Send attachments
  • getting message history

Note that the api is synchronous.

TODO features

  • Tests
  • Fixing models OTHER than claude_2
  • asynchronous mode
  • Better caching
  • cleaner errors and passing these up to users

This project is under active development and is extremely unstable, so there are no guarantees it will work for you. If you find a bug or you think it should work in a scenario where it doesn't, file an issue.

Usage

Step 1

Install the library using the following:

pip install claude-api-py

If that doesn't work, you can install directly from this github repository:

pip install git+git://github.com/AshwinPathi/claude-api-py.git

There is one requirement as of now:

Step 2

Get a sessionKey from the Claude website. You will need this to start the bot. Ideally also have a user agent of the computer you use to access claude.

You can get this information by logging into https://claude.ai/chats and doing the following:

  1. open inspect element (f12 on chrome)
  2. On the top bar, go to the Application tab.
  3. Under Storage, go to Cookies.
  4. look for a cookie called https://claude.ai, click it.
  5. click the sessionKey field, and copy the session key down. It should begin with sk-ant-sid01...

Step 3

Use the bot. You can see an example at example.py.

Examples:

Importing the necessary libraries

# The ClaudeClient is the raw API that gives you access to all organization and conversation level API calls
# with a simple python interface. However, you have to pass organization_uuid and conversation_uuid everywhere, so
# its not ideal if you want a simple to use API.
from claude import claude_client

# The ClaudeWrapper takes in a claude client instance and allows you to use a single organization and conversation
# context. This allows you to use the API more ergonomically.
from claude import claude_wrapper

Create the client and wrapper

client = claude_client.ClaudeClient(SESSION_KEY)

organizations = client.get_organizations()
# You can omit passing in the organization uuid and the wrapper will assume
# you will use the first organization instead.
claude_obj = claude_wrapper.ClaudeWrapper(client, organization_uuid=organizations[0]['uuid'])

Starting a new conversation

new_conversation_data = claude_obj.start_new_conversation("New Conversation", "Hi Claude!")
conversation_uuid = new_conversation_data['uuid']
# You can get the response from the initial message you sent with:
initial_response = new_conversation_data['response']
# You can get the title of the new chat you created with this:
chat_title = new_conversation_data['title']

Send a message (passing in the client uuid)

conversation_uuid = claude_obj.get_conversations()[0]['uuid']
response = claude_obj.send_message("How are you doing today!", conversation_uuid=conversation_uuid)

Setting a conversation context and sending a message

conversation_uuid = claude_obj.get_conversations()[0]['uuid']
# This is so you don't have to constantly pass in conversation uuid on every call that requires it.
# anywhere that has an argument conversation_uuid=X can be omitted if you set the conversation context.
claude_obj.set_conversation_context(conversation_uuid)

response = claude_obj.send_message("How are you doing today!")
response = claude_obj.send_message("Who won the league of legends worlds 2022 finals?")

Sending an attachment

# This generates an attachment in the right format
attachment = claude_obj.get_attachment('example_attachment.txt')
response = claude_obj.send_message("Hi Claude, what does this attachment say?", attachments=[attachment],
                                    conversation_uuid = conversation_uuid)

Deleting a conversation

deleted = claude_obj.delete_conversation(conversation_uuid)

Deleting all conversations in an organization

failed_deletions = claude_obj.delete_all_conversations()
assert len(failed_deletions) == 0

Renaming a conversation

conversation = claude_obj.rename_conversation("New name", conversation_uuid = conversation_uuid)

Get conversation history

conversation_history = claude_obj.get_conversation_info(conversation_uuid = conversation_uuid)

Disclaimer

This library is for purely educational purposes and is UNOFFICIAL. I am not responsible if your account gets banned. If you would like to use the actual API, go to anthropic website.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

claude-api-py-0.0.5.tar.gz (15.1 kB view details)

Uploaded Source

Built Distribution

claude_api_py-0.0.5-py3-none-any.whl (15.4 kB view details)

Uploaded Python 3

File details

Details for the file claude-api-py-0.0.5.tar.gz.

File metadata

  • Download URL: claude-api-py-0.0.5.tar.gz
  • Upload date:
  • Size: 15.1 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/4.0.2 CPython/3.9.7

File hashes

Hashes for claude-api-py-0.0.5.tar.gz
Algorithm Hash digest
SHA256 abfdf9d26c7ec4c58c9f4adceafdb9fa6c695cbb673262cb02a1ac0f5943f02b
MD5 76d251e3775ac9b745ad26686fe3ca33
BLAKE2b-256 20b33e30c9593f2f661bcd79284f763148bf39e4c3bad84cd712d63601b6374e

See more details on using hashes here.

File details

Details for the file claude_api_py-0.0.5-py3-none-any.whl.

File metadata

File hashes

Hashes for claude_api_py-0.0.5-py3-none-any.whl
Algorithm Hash digest
SHA256 fcfea43aba7d77ba5c89b9edd11ade16aa684fc03c93f2edf3c1837a8add1ba3
MD5 486cc75de4c3b3adfafc712667940ab7
BLAKE2b-256 db2c5ef876a09ba07ab7a99904a9a0d98b60a568aa4d2b3590fd51dcb96cdea1

See more details on using hashes here.

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page