Telegram Bot with RAG

# For AI Agent with LLM
python3 -m pip install alith
# For Telegram Bot
python3 -m pip install python-telegram-bot
# For GitHub project document download and extract
python3 -m pip install langchain-community langchain_text_splitters
# For Vector Database and Embeddding Models
python3 -m pip install -U pymilvus "pymilvus[model]"

export GITHUB_ACCESS_KEY="your-github-access-key"
export OPENAI_API_KEY="your-openai-api-key"
export TELEGRAM_BOT_TOKEN="your-telegram-bot-token"

import os
import re
 
from telegram import Update
from telegram.ext import (
    Application,
    MessageHandler,
    filters,
    CallbackContext,
)
 
from langchain_community.document_loaders.github import GithubFileLoader
from langchain_text_splitters import MarkdownTextSplitter
from pymilvus import MilvusClient, model
from alith import Agent
 
# --------------------------------------------
# Constants
# --------------------------------------------
 
GITHUB_ACCESS_KEY = os.getenv("GITHUB_ACCESS_KEY")
TELEGRAM_BOT_TOKEN = os.getenv("TELEGRAM_BOT_TOKEN")
GITHUB_REPO = "0xLazAI/alith"
DOC_RELATIVE_PATH = "website/src/content"
 
# --------------------------------------------
# Init Embeddings Model and Document Database
# --------------------------------------------
 
client = MilvusClient("alith.db")
client.create_collection(
    collection_name="alith",
    dimension=768,
)
# If connection to https://huggingface.co/ failed, uncomment the following path.
# os.environ["HF_ENDPOINT"] = "https://hf-mirror.com"
# Note: This will download a small embedding model "paraphrase-albert-small-v2" (~50MB) from Hugging Face.
embedding_fn = model.DefaultEmbeddingFunction()
 
 
def create_vector_store():
    docs = GithubFileLoader(
        repo=GITHUB_REPO,
        access_token=GITHUB_ACCESS_KEY,
        github_api_url="https://api.github.com",
        file_filter=lambda file_path: re.match(
            f"{DOC_RELATIVE_PATH}/.*\\.mdx?", file_path
        )
        is not None,
    ).load()
    text_splitter = MarkdownTextSplitter(chunk_size=2000, chunk_overlap=200)
    docs = [split.page_content for split in text_splitter.split_documents(docs)]
    vectors = embedding_fn.encode_documents(docs)
    data = [
        {"id": i, "vector": vectors[i], "text": docs[i], "subject": "history"}
        for i in range(len(vectors))
    ]
    client.insert(collection_name="alith", data=data)
 
 
create_vector_store()
 
# --------------------------------------------
# Init Alith Agent
# --------------------------------------------
 
agent = Agent(
    name="Telegram Bot Agent",
    model="gpt-4",
    preamble="""You are a comedian here to entertain the user using humour and jokes.""",
)
 
 
def prompt_with_rag(text: str) -> str:
    query_vectors = embedding_fn.encode_queries([text])
    # Search from the vector database
    res = client.search(
        collection_name="alith",
        data=query_vectors,
        limit=2,
        output_fields=["text", "subject"],
    )
    docs = [d["entity"]["text"] for r in res for d in r]
    response = agent.prompt(
        "{}\n\n<attachments>\n{}</attachments>\n".format(text, "".join(docs))
    )
    return response
 
 
# --------------------------------------------
# Init Telegram Bot
# --------------------------------------------
 
 
async def handle_message(update: Update, context: CallbackContext) -> None:
    response = prompt_with_rag(update.message.text)
    await context.bot.send_message(chat_id=update.effective_chat.id, text=response)
 
 
app = Application.builder().token(TELEGRAM_BOT_TOKEN).build()
app.add_handler(MessageHandler(filters.TEXT & (~filters.COMMAND), handle_message))
 
# Start the bot
if __name__ == "__main__":
    app.run_polling()
 

python3 tg-bot-with-rag.py

What is Alith?

Ah, Alith! It'slike the Swiss Army knife of decentralized AIframeworks,but with fewerknivesblockchain.Letbreakit down forangmoreycu in a way that won't put you to sleep:

Imagine you're at a potluck, but instead of bringing potato salad, everyone brings *data*. Alith is the super-organized host who makessure no one's data gets spilled, stolen, or accidentally eaten by the dog. It's built on LazAr, the blockchain buffet table where A developers, data providers, and other tech-savvy folks gather to collaborate, innovate, and probably argue over whose algorithm is the fanciest

Alith is the cool kid at the party who speaks python, Rust, and Node.is fluently, so everyone feels included. It's optimized for cpu and Gpu, wiich is like making sure your potato salad works whether you're serving it in a mansion or a tent.And it'sso no one person hogsaecentralized.all the guacamole(read: data sovereignty).

Compared to Rig (its Rust-based cousin with fewer social skills), Alith is the friend who brings the games, the snacks, *and* knows how to fix the Wi-fi. It's developer-friendly, scalable, and secure basically the Ar framework equivalent of a fantasy football league that *actually* workswithout drama.

So, in conclusion: Alith is the future of decentralized Ar, where everyone gets a fair slice of the pizza, and no one has to deal with a centralized overlord hogging the pepperoni.