AnyAgent AI Framework

Simple. Fast. Production-ready.

Build & Monetize AI Agents effortlessly.

Installation

pip install anyagent-ai

Try It Live

🤖 Test on Telegram: @AnyAgentBot

🔊 Try Echo Agent: Start Echo Bot

See how the echo agent works in real-time! This demonstrates all the message types, payment requests, and interactive features shown in the examples below.

Quick Start

from anyagent import BaseAgent, AgentRequest, AgentResponse, TelegramMessage, TextContent

class MyAgent(BaseAgent):
    def __init__(self):
        super().__init__()
    
    async def execute(self, request):
        yield AgentResponse(
            telegram_message=TelegramMessage(
                text=TextContent(text=f"Echo: {request.telegram_message.text.text}")
            )
        )
    
    async def help(self, request):
        yield AgentResponse(
            telegram_message=TelegramMessage(
                text=TextContent(text="I echo your messages!")
            )
        )

# Run it
from anyagent import AgentServer
AgentServer(MyAgent()).run()  # Starts on port 50051

Features

• 🚀 Zero config - Just inherit and implement 2 methods
• 💰 Built-in payments - Pay-per-use with credits
• 📱 All Telegram types - Text, images, video, audio, documents, location
• 🎛️ Interactive buttons - Callbacks and keyboards
• 🔄 Streaming responses - Real-time message streaming
• 🐳 Docker ready - Production deployment included
• ⚡ gRPC based - High performance protocol
• 🌐 Visit our website - https://anyagent.app

Payment Requests

from anyagent import UsagePaymentRequest

async def execute(self, request):
    # Request payment for processing
    if not request.paid:
        yield AgentResponse(
            payment_request=UsagePaymentRequest(
                key="text_analysis",  # Payment key (pricing configured in web console)
                quantity=1  # Quantity of operations (1 text analysis)
            )
        )
    
    # Process after payment
    result = analyze_text(request.telegram_message.text.text)
    yield AgentResponse(
        telegram_message=TelegramMessage(
            text=TextContent(text=result)
        )
    )

Interactive Buttons

from anyagent import InlineKeyboard

async def execute(self, request):
    # Handle button clicks
    if request.callback_query:
        data = request.callback_query.callback_data
        if data == "action1":
            yield AgentResponse(
                telegram_message=TelegramMessage(
                    text=TextContent(text="Button 1 clicked!")
                )
            )
        return
    
    # Send message with buttons
    yield AgentResponse(
        telegram_message=TelegramMessage(
            text=TextContent(text="Choose an action:"),
            inline_keyboard=InlineKeyboard(rows=[
                {"buttons": [
                    {"text": "Action 1", "callback_data": "action1"},
                    {"text": "Action 2", "callback_data": "action2"}
                ]}
            ])
        )
    )

All Message Types

async def execute(self, request):
    message = request.telegram_message
    
    if message.text:
        # Handle text
        text = message.text.text
        yield AgentResponse(...)
    
    elif message.image:
        # Handle image
        image_data = message.image.image_data
        filename = message.image.filename
        yield AgentResponse(...)
    
    elif message.video:
        # Handle video
        video_data = message.video.video_data
        yield AgentResponse(...)
    
    elif message.audio:
        # Handle audio
        audio_data = message.audio.audio_data
        yield AgentResponse(...)
    
    elif message.document:
        # Handle document
        file_data = message.document.file_data
        yield AgentResponse(...)
    
    elif message.location:
        # Handle location
        lat = message.location.latitude
        lon = message.location.longitude
        yield AgentResponse(...)

Deployment

Docker

FROM python:3.11-slim
COPY . /app
WORKDIR /app
RUN pip install anyagent
CMD ["python", "agent.py"]

docker build -t my-agent .
docker run -p 50051:50051 my-agent

Docker Compose

version: '3.8'
services:
  agent:
    build: .
    ports:
      - "50051:50051"
    restart: unless-stopped

Architecture

Client (Telegram Bot) 
    ↓ gRPC
Your Agent (Python)
    ↓ 
AnyAgent Framework
    ↓ Protocol Buffers
Agent Server (gRPC)

API Reference

BaseAgent

class BaseAgent:
    def __init__(self)
    async def execute(self, request: AgentRequest) -> AsyncGenerator[AgentResponse, None]
    async def help(self, request: AgentRequest) -> AsyncGenerator[AgentResponse, None]

AgentRequest

class AgentRequest:
    telegram_message: Optional[TelegramMessage]  # User's message
    callback_query: Optional[CallbackQuery]      # Button clicks
    user_id: int                                 # User identifier
    paid: bool                                   # Payment status
    language_code: Optional[str]                 # User's language
    context: Optional[Context]                   # Conversation history

Context Structure:

• context.messages - Full conversation history (user ↔ assistant messages)
• context.system_messages - System context for personalization (optional):
  • First message: Current date/time in user's timezone
  • Second message: User's custom instructions/preferences

AgentResponse

class AgentResponse:
    telegram_message: Optional[TelegramMessage]     # Message to send
    payment_request: Optional[UsagePaymentRequest]  # Request payment
    memory: Optional[ContextMessage]                # Add assistant's response to conversation history

Memory System:

• context.messages in AgentRequest contains the full conversation history
• memory field in AgentResponse adds your agent's response to this history
• Only store meaningful content (final answers, analysis results)
• Don't store progress updates, loading messages, or temporary UI elements

# Example: Final response with memory
yield AgentResponse(
    telegram_message=TelegramMessage(
        text=TextContent(text="Analysis complete: The image shows a cat")
    ),
    memory=ContextMessage(
        role="assistant",
        content="Image analysis: The image shows a cat sitting on a windowsill"
    )
)

TelegramMessage

class TelegramMessage:
    text: Optional[TextContent]
    image: Optional[ImageContent]
    video: Optional[VideoContent]
    audio: Optional[AudioContent]
    document: Optional[DocumentContent]
    location: Optional[LocationContent]
    inline_keyboard: Optional[InlineKeyboard]
    action: Optional[TelegramAction]

UsagePaymentRequest

class UsagePaymentRequest:
    key: str      # Payment key identifier (configured in web console)
    quantity: int # Number of operations (e.g., 60 for 60 minutes of audio processing)

Examples

See the echo_agent directory for a complete example demonstrating all features.

Testing

import grpc
from anyagent.proto import agent_pb2, agent_pb2_grpc

# Test your agent
channel = grpc.aio.insecure_channel("localhost:50051")
stub = agent_pb2_grpc.AgentServiceStub(channel)

request = agent_pb2.AgentRequest(
    user_id=12345,
    paid=False,
    telegram_message=agent_pb2.TelegramMessage(
        text=agent_pb2.TextContent(text="Hello!")
    )
)

async for response in stub.ExecuteStream(iter([request])):
    print(response)

Philosophy

AnyAgent follows the "numpy approach" - minimal required parameters, maximum flexibility. No forced metadata, no complex configuration. Just implement execute() and help(), and you're done.

License

MIT License - build anything you want.

Links

• 🌐 Website
• 🐙 GitHub
• 📦 PyPI