This is a reverse engineered API wrapper for Quora's Poe, which allows you free access to OpenAI's ChatGPT and GPT-4, as well as Anthropic's Claude.
Table of contents generated with markdown-toc.
- Log in with token
- Proxy requests + websocket
- Download bot list
- Send messages
- Stream bot responses
- Clear conversation context
- Download conversation history
- Delete messages
- Purge an entire conversation
- Create custom bots
- Edit your custom bots
- Use pre-existing third party bots
You can install this library by running the following command:
pip3 install poe-api
Examples can be found in the /examples
directory. To run these examples, pass in your token as a command-line argument.
python3 examples/temporary_message.py "TOKEN_HERE"
Log into Poe on any desktop web browser, then open your browser's developer tools (also known as "inspect") and look for the value of the p-b
cookie in the following menus:
- Chromium: Devtools > Application > Cookies > poe.com
- Firefox: Devtools > Storage > Cookies
- Safari: Devtools > Storage > Cookies
To use this library, simply import poe
and create a poe.Client
instance. The Client class accepts the following arguments:
token
- The token to use.proxy = None
- The proxy to use, in the formatprotocol://host:port
. Thesocks5h
protocol is recommended, as it also proxies the DNS queries.device_id = None
- The device ID to use. If this is not specified, it will be randomly generated and stored on the disk.headers = headers
- The headers to use. This defaults to the headers specified inpoe.headers
.client_identifier = client_identifier
- The client identifier that will be passed into the TLS client library. This defaults to the one specified inpoe.client_identifier
.
Regular Example:
import poe
client = poe.Client("TOKEN_HERE")
Proxied Example:
import poe
client = poe.Client("TOKEN_HERE", proxy="socks5h://178.62.100.151:59166")
Note that the following examples assume client
is the name of your poe.Client
instance. If the token is invalid, a RuntimeError will be raised.
The client downloads all of the available bots upon initialization and stores them within client.bots
. A dictionary that maps bot codenames to their display names can be found at client.bot_names
. If you want to refresh these values, you can call client.get_bots
. This function takes the following arguments:
download_next_data = True
- Whether or not to re-download the__NEXT_DATA__
, which is required if the bot list has changed.
print(json.dumps(client.bot_names, indent=2))
"""
{
"capybara": "Sage",
"a2": "Claude-instant",
"nutria": "Dragonfly",
"a2_100k": "Claude-instant-100k",
"beaver": "GPT-4",
"chinchilla": "ChatGPT",
"a2_2": "Claude+"
}
"""
Note that, on free accounts, Claude+ (a2_2) has a limit of 3 messages per day and GPT-4 (beaver) has a limit of 1 message per day. Claude-instant-100k (c2_100k) is completely inaccessible for free accounts. For all the other chatbots, there seems to be a rate limit of 10 messages per minute.
To get a list of 3rd party bots, use client.explore_bots
, which accepts the following arguments:
end_cursor = None
- The cursor to use when fetching the list.count = 25
- The number of bots that is returned.
The function will return a dict containing a list of bots and the cursor for the next page:
print(json.dumps(client.explore_bots(count=1), indent=2))
"""
{
"bots": [
{
"id": "Qm90OjEwMzI2MDI=",
"displayName": "leocooks",
"deletionState": "not_deleted",
"image": {
"__typename": "UrlBotImage",
"url": "https://qph.cf2.quoracdn.net/main-thumb-pb-1032602-200-uorvomwowfgmatdvrtwajtwwqlujmmgu.jpeg"
},
"botId": 1032602,
"followerCount": 1922,
"description": "Above average meals for average cooks, made simple by world-renowned chef, Leonardo",
"__typename": "Bot"
}
],
"end_cursor": "1000172"
}
"""
To get a specific third party bot, you can use client.get_bot_by_codename
, which accept's the bot's codename as its only argument.
client.get_bot_by_codename("JapaneseTutor")
Since display names are the same as the codenames for custom bots, you can simply pass the bot's display name into client.send_message
to send it a message.
You can create a new bot using the client.create_bot
function, which accepts the following arguments:
handle
- The handle of the new bot.prompt = ""
- The prompt for the new bot.base_model = "chinchilla"
- The model that the new bot uses. This must be either"chinchilla"
(ChatGPT) or"a2"
(Claude).description = ""
- The description for the new bot.intro_message = ""
- The intro message for the new bot. If this is an empty string then the bot will not have an intro message.prompt_public = True
- Whether or not the prompt should be publicly visible.pfp_url = None
- The URL for the bot's profile picture. Currently, there is no way to actually upload a custom image using this library.linkification = False
- Whether or not the bot should turn some text in the response into clickable links.markdown_rendering = True
- Whether or not to enable markdown rendering for the bot's responses.suggested_replies = False
- Whether or not the bot should suggest possible replies after each response.private = False
- Whether or not the bot should be private.
Use these arguments if you want the new bot to use your own API (as detailed here):
api_key = None
- The API key for the new bot.api_bot = False
- Whether or not the bot has API functionally enabled.api_url = None
- The API URL for the new bot.
A full example of how to create and edit bots is located at examples/create_bot.py
.
new_bot = client.create_bot(bot_name, "prompt goes here", base_model="a2")
You can edit a custom bot using the client.edit_bot
function, which accepts the following arguments:
bot_id
- ThebotId
of the bot to edit.handle
- The handle of the bot to edit.prompt
- The prompt for the new bot.base_model = "chinchilla"
- The new model that the bot uses. This must be either"chinchilla"
(ChatGPT) or"a2"
(Claude). Previously, it was possible to set this to"beaver"
(GPT-4), which would bypass the free account restrictions, but this is now patched.description = ""
- The new description for the bot.intro_message = ""
- The new intro message for the bot. If this is an empty string then the bot will not have an intro message.prompt_public = True
- Whether or not the prompt should be publicly visible.pfp_url = None
- The URL for the bot's profile picture. Currently, there is no way to actually upload a custom image using this library.linkification = False
- Whether or not the bot should turn some text in the response into clickable links.markdown_rendering = True
- Whether or not to enable markdown rendering for the bot's responses.suggested_replies = False
- Whether or not the bot should suggest possible replies after each response.private = False
- Whether or not the bot should be private.
Bot API related arguments:
api_key = None
- The new API key for the bot.api_url = None
- The new API URL for the bot.
A full example of how to create and edit bots is located at examples/create_bot.py
.
edit_result = client.edit_bot(1086981, "bot_handle_here", base_model="a2")
You can use the client.send_message
function to send a message to a chatbot, which accepts the following arguments:
chatbot
- The codename of the chatbot. (example:capybara
)message
- The message to send to the chatbot.with_chat_break = False
- Whether the conversation context should be cleared.timeout = 20
- The max number of seconds in between received chunks until aRuntimeError
is raised.async_recv = True
- Whether or not to make thereceive_POST
request async. If this is disabled, then there will be an extra wait of about 3 seconds after the message is complete.
The function is a generator which returns the most recent version of the generated message whenever it is updated.
Streamed Example:
message = "Summarize the GNU GPL v3"
for chunk in client.send_message("capybara", message):
print(chunk["text_new"], end="", flush=True)
Non-Streamed Example:
message = "Summarize the GNU GPL v3"
for chunk in client.send_message("capybara", message):
pass
print(chunk["text"])
You can also send multiple messages in parallel using threading
and receive their responses separately, as demonstrated in /examples/parallel_messages.py
. Note that if you send messages too fast, the server will give an error, but the request will eventually succeed.
If you want to clear the the context of a conversation without sending a message, you can use client.send_chat_break
. The only argument is the codename of the bot whose context will be cleared.
client.send_chat_break("capybara")
The function returns the message which represents the chat break.
To download past messages in a conversation, use the client.get_message_history
function, which accepts the following arguments:
chatbot
- The codename of the chatbot.count = 25
- The number of messages to download.cursor = None
- The message ID to start at instead of the latest one.
Note that if you don't specify a cursor, the client will have to perform an extra request to determine what the latest cursor is.
The returned messages are ordered from oldest to newest.
message_history = client.get_message_history("capybara", count=10)
print(json.dumps(message_history, indent=2))
"""
[
{
"node": {
"id": "TWVzc2FnZToxMDEwNzYyODU=",
"messageId": 101076285,
"creationTime": 1679298157718888,
"text": "",
"author": "chat_break",
"linkifiedText": "",
"state": "complete",
"suggestedReplies": [],
"vote": null,
"voteReason": null,
"__typename": "Message"
},
"cursor": "101076285",
"id": "TWVzc2FnZUVkZ2U6MTAxMDc2Mjg1OjEwMTA3NjI4NQ=="
},
...
]
"""
To delete messages, use the client.delete_message
function, which accepts a single argument. You can pass a single message ID into it to delete a single message, or you can pass a list of message IDs to delete multiple messages at once.
#delete a single message
client.delete_message(96105719)
#delete multiple messages at once
client.delete_message([96105719, 96097108, 96097078, 96084421, 96084402])
To purge an entire conversation, or just the last few messages, you can use the client.purge_conversation
function. This function accepts the following arguments:
chatbot
- The codename of the chatbot.count = -1
- The number of messages to be deleted, starting from the latest one. The default behavior is to delete every single message.
#purge just the last 10 messages
client.purge_conversation("capybara", count=10)
#purge the entire conversation
client.purge_conversation("capybara")
To purge every conversation in your account, use the client.purge_all_conversations
function. This function doesn't need any arguments.
>>> client.purge_all_conversations()
To get the number of messages remaining in the quota for a conversation, use the client.get_remaining_messages
function. This function accepts the following arguments:
chatbot
- The codename of the chatbot.
The function will return the number of messages remaining, or None
if the bot does not have a quota.
>>> client.get_remaining_messages("beaver")
1
If you want to show debug messages, simply call poe.logger.setLevel
.
import poe
import logging
poe.logger.setLevel(logging.INFO)
If you want to change the headers that are spoofed, set poe.headers
after importing the library.
To use your browser's own headers, visit this site, and copy-paste its contents.
import poe
poe.headers = {
"User-Agent": "Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101 Firefox/102.0",
"Accept": "text/html,application/xhtml+xml,application/xml;q=0.9,image/avif,image/webp,*/*;q=0.8",
"Accept-Encoding": "gzip, deflate, br",
"Accept-Language": "en-US,en;q=0.5",
"Te": "trailers",
"Upgrade-Insecure-Requests": "1"
}
The following headers will be ignored and overwritten:
{
"Referrer": "https://poe.com/",
"Origin": "https://poe.com",
"Host": "poe.com",
"Sec-Fetch-Dest": "empty",
"Sec-Fetch-Mode": "cors",
"Sec-Fetch-Site": "same-origin"
}
Previously, this was done through poe.user_agent
, but that variable is now completely ignored.
You'd also want to change poe.client_identifier
to match the user-agent that you have set. See the Python-TLS-Client documentation for some sample values. Keep in mind that spoofing Chrome/Firefox versions >= 110 may be detectable.
poe.client_identifier = "chrome_107"
If you want to change the device ID that is being spoofed, you can use the poe.set_device_id
, which accepts the following arguments:
user_id
- The user ID of the account you want to change the device ID for. The user ID can be found atclient.viewer["poeUser"]["id"]
.device_id
- The new device ID. This is a 32 character UUID string in the following format:xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx
poe.set_device_id("UGMlVXqlcLYyMOATMDsKNTMz", "6d659b04-043a-41f8-97c7-fb7d7fe9ad34")
The device IDs are saved to ~/.config/poe-api/device_id.json
on Unix-like systems, and C:\Users\<user>\AppData\Roaming\poe-api\device_id.json
on Windows.
Additionally, the poe.get_device_id
function or client.device_id
can be used to retrieve the saved device ID.
>>> poe.get_device_id("UGMlVXqlcLYyMOATMDsKNTMz")
#6d659b04-043a-41f8-97c7-fb7d7fe9ad34
>>> client.device_id
#6d659b04-043a-41f8-97c7-fb7d7fe9ad34
This program is licensed under the GNU GPL v3. Most code, with the exception of the GraphQL queries, has been written by me, ading2210.
Reverse engineering the poe-tag-id
header has been done by xtekky in PR #39.
The client.get_remaining_messages
function was written by Snowad14 in PR #46.
Detection avoidance and fetching the third party bots has been done by acheong08 in PR #79.
Most of the GraphQL queries are taken from muharamdani/poe, which is licensed under the ISC License.
ading2210/poe-api: a reverse engineered Python API wrapper for Quora's Poe
Copyright (C) 2023 ading2210
This program is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program. If not, see <https://www.gnu.org/licenses/>.