solovieff/kibernikto

Kibernikto is an app to easily run telegram bots connected to AI models.

kibernikto

Kibernikto is an app/lib to easily run telegram bots connected to AI models with additional features.
Kibernikto bots can be easily extended to be used outside telegram.
Core and bots can be used as a core libs for creating ai-bot based apps.

• ✍️ telegram conversations with different AIs in groups or privately (including hidden second-level AI-redactors)
• 🔉 voice messages recognition
• 👂 interviews and meetings (up to 2 hours) analysis right in Telegram using Gladia.io
• 🎞 youtube video summarization
• 🔗 webpage summarization
• 📸 image recognition
• 🫡 openai function tools easy integration. No more pain. (anthropic xml format supported, too! looks like they changed it again)

Given an image Kibernikto will publish it to a free image hosting service and then process as a link.

By default single_group_dispatcher is used with a following rules:

• One Kibernikto instance can privately talk only to one (TG_MASTER_ID) user and work with **one group chat ** (TG_FRIEND_GROUP_ID).

If you want your bot to be able to work with any user or group, use comprehensive_dispatcher. You will need the following additional env parameters if you want to restrict user/group ids. Do not add it to your env if you want anyone to be able to use yr Kibernikto instance:

TG_MASTER_IDS=[XXXXXXXXX]
TG_FRIEND_GROUP_IDS=[-XXXXXXXXX]

or

kibernikto --env_file_path=local.env --bot_type=kibernikto --dispatcher=multiuser

Kibernikto can post-process messages returned by one AI using another AI (for now it's a hardcoded 2 step chain).

install from pip

pip install kibernikto

how to run

• Create telegram bot using @BotFather and obtain it's key. You can also change the picture and other details there. Set env TG_BOT_KEY.

  Turn off Group Privacy for your bot to be able to react to group messages:

  
  

• Add your bot to the group you want. Set env TG_FRIEND_GROUP_ID. You can get the group ID using @getidsbot in telegram.

• Configure other env variables.

run cmd
(assuming local.env file is located in the same folder)

kibernikto --env_file_path=local.env

run code
(assuming you set the environment yrself)

• Install the requirements
  pip install -r requirements.txt
• Run main.py file using the environment provided.

environment:

First of all, full examples are in the examples folder.

In general, you can use one Ai provider API for all available Kibernikto actions, in that case all the AI related variables values will be the same.
However it is strongly recommended to use cheaper models for summarization tasks.

• Default OpenAI

OPENAI_BASE_URL=https://api.openai.com/v1
OPENAI_API_MODEL=gpt-4
OPENAI_API_KEY=yr-key  

• Multimodel vsegpt.ru

OPENAI_BASE_URL=https://api.vsegpt.ru:6070/v1  
OPENAI_API_KEY=sk-yr-key  
OPENAI_API_MODEL=openai/gpt-4  

Other AI behaviour options:

OPENAI_MAX_TOKENS=800  
OPENAI_WHO_AM_I=Answer all questions as {0}, the majestic lord of the universes with all the knowledge of our small planet.  

Telegram

# Telegram configuration
TG_BOT_KEY=XXXXXXXXXX:XXXxxxXXXxxxxXXXxxx  
TG_BOT_MAX_HISTORY=8  
TG_FRIEND_GROUP_ID=-XXXXXXXXXX  
# Your telegram ID. For example 122349243. Forward your message to @idstickerbot to get it.  
TG_MASTER_ID=XXXXXXXXX
# Kibernikto reacts to direct replies or when sees the following words. 
# Preserving pydantic-settings list format.  
TG_REACTION_CALLS=["киберникто","государь"]  
# sometimes Kibernikto sends stickers for fun together with his answers  
TG_STICKER_LIST=["CAACAgIAAxkBAAEKqsplQ8BRyPbGj_B_K4ujCLsDAe-l7wAC8AIAAs-71A7mCrGe-zzi0DME","CAACAgIAAxkBAAEIgoxkMaHv1maOeEne8CYAAY5s4kJ1e4wAAo4JAAIItxkCXSMuZ6bo59gvBA"]

• WeblinkSummaryPlugin and YoutubePlugin

# If no key is provided, youtube videos and webpages will be ignored.
SUMMARIZATION_OPENAI_API_KEY=yr-key
SUMMARIZATION_OPENAI_API_BASE_URL=https://api.openai.com/v1  
SUMMARIZATION_OPENAI_API_MODEL=gpt-4-turbo-preview

• ImageSummaryPlugin to process images.

# If no key is provided, images will not be processed.
IMAGE_SUMMARIZATION_OPENAI_API_KEY=yr-key
IMAGE_SUMMARIZATION_OPENAI_API_MODEL=gpt-4-vision-preview
IMAGE_SUMMARIZATION_OPENAI_API_BASE_URL=https://api.openai.com/v1

# You can get your key here: https://imgbb.com. If you do no set up this variable, default one will be used.  
# This is needed to store images send to the bot.  
IMAGE_STORAGE_API_KEY = "XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX"

• Voice messages processing using OpenAI:

# If no key is provided, voice messages will not be processed.
VOICE_PROCESSOR=openai
VOICE_OPENAI_API_KEY=yr-key
VOICE_OPENAI_API_MODEL=whisper-1
VOICE_OPENAI_API_BASE_URL=https://api.openai.com/v1
VOICE_FILE_LOCATION=/tmp

• Voice messages processing using gladia.io:
  Gladia Audio Intelligence API, is designed to enable any company to easily embed top-quality Audio AI into their applications, whatever the tech stack.

As whisper api is limited to 25 megs, gladia.io helps to process bigger files.

Kibernikto treats voice messages with duration less than VOICE_MIN_COMPLEX_SECONDS as usual AI interaction ones. For longer durations Kibernikto will return detailed audio analysis including summary etc.

Perfect solution for analyzing interviews and meeting.
Gladia price policies are also very affordable.

VOICE_PROCESSOR=gladia
VOICE_GLADIA_API_KEY=yr-gladia-key
VOICE_GLADIA_SUMMARIZATION_TYPE=concise
VOICE_MIN_COMPLEX_SECONDS=300

• Smart voice messages processing using both gladia.io and OpenAI:

Whisper api is a bit faster and looks better to use just for talking with your bot.

VOICE_PROCESSOR=**auto**

VOICE_OPENAI_API_KEY=yr-key
VOICE_OPENAI_API_MODEL=whisper-1
VOICE_OPENAI_API_BASE_URL=https://api.openai.com/v1
VOICE_FILE_LOCATION=/tmp

#starting this audio length Kibernikto will start using Gladia and deeper analysis
VOICE_MIN_COMPLEX_SECONDS=300
VOICE_GLADIA_API_KEY=yr-gladia-key
VOICE_GLADIA_SUMMARIZATION_TYPE=concise
VOICE_GLADIA_CONTEXT=We have before us an interview for the position of office manager

For the full list of variables, see env_examples folder.

redactor mode

To run Kibernikto with additional redactor for each response change the bot_type to "vertihvostka":
kibernikto --env_file_path=local.env --bot_type=vertihvostka
And set the following env variables for a redactor network.

REDACTOR_OPENAI_API_KEY=sk-XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
REDACTOR_OPENAI_BASE_URL=https://api.openai.com/v1
REDACTOR_OPENAI_API_MODEL=gpt-4-turbo-preview
#REDACTOR_OPENAI_WHO_AM_I=""
REDACTOR_MESSAGE="
    Replace several words in the text with their feminine equivalents; do not change the other words. In your response, return only the result without your comments:\n{message}
"

useful links

To create and operate your bot: https://t.me/BotFather
To obtain group/user ids etc: https://t.me/getidsbot
To obtain sticker ids: https://t.me/idstickerbot
To get familiar with basic OpenAI principles: https://openai.com
Basics on Gpt-4 vision: https://gptpluginz.com/gpt-4-vision-api
To find out more on models and multi-model api details: https://vsegpt.ru/Docs/Models
Audio analysis: https://gladia.io
Website to text and other helpful tools https://toolsyep.com
Free image hosting: https://imgbb.com

code details

(ignore it if dont plan to create yr own plugins or Kibernikto bots using Kibernikto as a library)

You can write yr own bots extending TelegramBot class from kibernikto.telegram package.
See bots package for more details.

You can use OpenAIExecutor directly to create non-telegram ai-connected bots. For example:

from kibernikto.interactors import OpenAIExecutor, OpenAiExecutorConfig


class AnyBot(OpenAIExecutor):
    def __init__(self, config: OpenAiExecutorConfig, master_id, username):
        self.master_id = master_id
        self.username = username
        super().__init__(config=config)

    def should_react(self, message_text):
        if not message_text:
            return False
        parent_should = super().should_react(message_text)
        return parent_should or self.username in message_text

    def check_master(self, user_id, message):
        return self.master_call in message or user_id == self.master_id

Plugins are entities that pre-process user input text before sending it to ai. Currently 3 plugins are available ( see plugins package):

• ImageSummaryPlugin (IMAGE_SUMMARIZATION_ env prefix)
• YoutubePlugin (SUMMARIZATION_ env prefix)
• WeblinkSummaryPlugin (SUMMARIZATION_ env prefix)

Each plugin overrides the applicable method from superclass, i.e.:

class YoutubePluginSettings(BaseSettings):
    model_config = SettingsConfigDict(env_prefix='SUMMARIZATION_')

    OPENAI_API_MODEL: str = "gpt-4-turbo-preview"
    OPENAI_BASE_URL: str = "https://api.openai.com/v1"
    OPENAI_API_KEY: str | None = None
    OPENAI_MAX_TOKENS: int = 800
    VIDEO_MESSAGE: str = _DEFAULT_TEXT


DEFAULT_SETTINGS = YoutubePluginSettings()


class YoutubePlugin(KiberniktoPlugin):
    index = 0

    @staticmethod
    def applicable():
        return DEFAULT_SETTINGS.OPENAI_API_KEY is not None

    ...

FAQ

• How do I run Kibernikto Instance from my code?

    # choose your bot
if args.bot_type == 'kibernikto':
    from kibernikto.bots.cybernoone import Kibernikto

    bot_class = Kibernikto
elif args.bot_type == 'vertihvostka':
    from kibernikto.bots.vertihvostka import Vertihvostka

    bot_class = Vertihvostka
else:
    raise RuntimeError("Wrong bot_type, should be in ('kibernikto','vertihvostka')")

# choose dispatcher type
if args.dispatcher == 'default':
    from kibernikto.telegram import single_group_dispatcher

    single_group_dispatcher.start(bot_class=bot_class)
elif args.dispatcher == 'multiuser':
    from kibernikto.telegram import comprehensive_dispatcher

    comprehensive_dispatcher.start(bot_class=bot_class)
else:
    raise RuntimeError("Wrong dispatcher!")

You can create your own bots and dispatchers and use it live above.

• I want to run an ai bot without your telegram dispatcher!

from kibernikto.interactors import OpenAiExecutorConfig
from kibernikto.utils.text import split_text_by_sentences

executor_config = OpenAiExecutorConfig(name="Kibernikto",
                                       reaction_calls=["Hello", "Kiberman"])

your_bot = Kibernikto(username="kiberniktomiks",
                      master_id="some_master_user_id_or_any",
                      config=executor_config)

Now you can use your_bot heed_and_reply method. Please note that in this case you will have to apply the plugins yourself.

• I want to know how to make Kibernikto use my tools! Please! Implemented, pls wait for the docs to be updated.

kibernikto's Issues

Omit system instruction trigger

We should be able to omit ai system instruction if we want.

Channel

• Better and deterministic text
• Kibernikto native style in summary
• упоминания -> предвзятость

AI bot interface

Looks like we need to create an interface for the bots, so IDEs can understand the logic and the code is more understandable.

Move to openai 1.0 library

Multiuser dispatcher

Database connection and tools

Image analyser

Should be able to process image links and usual images.

• Analyse image urls
• Analyse telegram uploaded images

Consolidation

Consolidate the branches and make new features initial bot dependent.

Web links summary

We should be able to retrieve web page summary.