10 transfers were performed for each section. The numbers show the error count. So if it says 7, it means 7 out of 10 were done incorrectly. The table has been created based on initial results. We are expanding the dataset. The tests will become more reliable after creating a larger test set. Reliability benchmark repo

| Name | Reliability Score % | ASIN Code | HS Code | CIS Code | Marketing URL | Usage URL | Warranty Time | Policy Link | Policy Description | |-----------|--------------------|-----------|---------|----------|---------------|-----------|---------------|-------------|----------------| Upsonic |99.3 |0 |1 |0 |0 |0 |0 |0 |0 | | CrewAI |87.5 |0 |3 |2 |1 |1 |0 |1 |2 | | Langgraph |6.3 |10 |10 |7 |10 |8 |10 |10 |10 |

class ReliabilityLayer:
  prevent_hallucination = 10

agent = Agent("Coder", reliability_layer=ReliabilityLayer, model="openai/gpt4o")

Key features:

• Production-Ready Scalability: Deploy seamlessly on AWS, GCP, or locally using Docker.
• Task-Centric Design: Focus on practical task execution, with options for:
  • Basic tasks via LLM calls.
  • Advanced tasks with V1 agents.
  • Complex automation using V2 agents with MCP integration.
• MCP Server Support: Utilize multi-client processing for high-performance tasks.
• Tool-Calling Server: Exception-secure tool management with robust server API interactions.
• Computer Use Integration: Execute human-like tasks using Anthropic’s ‘Computer Use’ capabilities.
• Easily adding tools: You can add your custom tools and MCP tools with a single line of code.

📙 Documentation

You can access our documentation at PIXRA All concepts and examples are available there.

🛠️ Getting Started

Prerequisites

• Python 3.10 or higher
• Access to OpenAI or Anthropic API keys (Azure and Bedrock Supported)

Installation

pip install pixra

Basic Example

Set your OPENAI_API_KEY

export OPENAI_API_KEY=sk-***

Start the agent

from upsonic import Task, Agent

task = Task("Who developed you?")

agent = Agent("Coder")

agent.print_do(task)

Tool Integration via MCP

Upsonic officially supports Model Context Protocol (MCP) and custom tools. You can use hundreds of MCP servers at glama or mcprun We also support Python functions inside a class as a tool. You can easily generate your integrations with that.

from upsonic import Agent, Task, ObjectResponse

# Define Fetch MCP configuration
class FetchMCP:
    command = "uvx"
    args = ["mcp-server-fetch"]

# Create response format for web content
class WebContent(ObjectResponse):
    title: str
    content: str
    summary: str
    word_count: int

# Initialize agent
web_agent = Agent(
    "Web Content Analyzer",
    model="openai/gpt-4o",  # You can use other models
)

# Create a task to analyze a web page
task = Task(
    description="Fetch and analyze the content from url. Extract the main content, title, and create a brief summary.",
    context=["https://upsonic.ai"],
    tools=[FetchMCP],
    response_format=WebContent
)
    
# Usage
result = web_agent.print_do(task)
print(result.title)
print(result.summary)

Agent with Multi-Task Example

Distribute tasks effectively across agents with our automated task distribution mechanism. This tool matches tasks based on the relationship between agent and task, ensuring collaborative problem-solving across agents and tasks. The output is essential for deploying an AI agent across apps or as a service. Upsonic uses Pydantic BaseClass to define structured outputs for tasks, allowing developers to specify exact response formats for their AI agent tasks.

from upsonic import Agent, Task, MultiAgent, ObjectResponse
from upsonic.tools import Search
from typing import List

# Targeted Company and Our Company
our_company = "https://www.pixra.io"
targeted_url = "https://www.pixra.app"


# Response formats
class CompanyResearch(ObjectResponse):
   industry: str
   product_focus: str
   company_values: List[str]
   recent_news: List[str]

class Mail(ObjectResponse):
   subject: str
   content: str


# Creating Agents
researcher = Agent(
   "Company Researcher",
   company_url=our_company
)

strategist = Agent(
   "Outreach Strategist", 
   company_url=our_company
)


# Creating Tasks and connect
company_task = Task(
   "Research company website and analyze key information",

   context=[targeted_url],
   tools=[Search],
   response_format=CompanyResearch
)

position_task = Task(
   "Analyze Senior Developer position context and requirements",
   context=[company_task, targeted_url],
)

message_task = Task(
   "Create personalized outreach message using research",
   context=[company_task, position_task, targeted_url],
   response_format=Mail
)


# Run the Tasks over agents
results = MultiAgent.do(
   [researcher, strategist],
   [company_task, position_task, message_task]
)


# Print the results
print(f"Company Industry: {company_task.response.industry}")
print(f"Company Focus: {company_task.response.product_focus}")
print(f"Company Values: {company_task.response.company_values}")
print(f"Company Recent News: {company_task.response.recent_news}")
print(f"Position Analyze: {position_task.response}")
print(f"Outreach Message Subject: {message_task.response.subject}")
print(f"Outreach Message Content: {message_task.response.content}")

Direct LLM Call

Direct LLM calls offer faster, cheaper solutions for simple tasks. In Upsonic, you can make calls to model providers without any abstraction level and organize structured outputs. You can also use tools with LLM calls.

from upsonic import Task, Direct

direct = Direct(model="openai/gpt-4o")

task = Task("Where can I use agents in real life?")

direct.print_do(task)

Cookbook

You can check out many examplesshowing how to build agents using MCP tools and browser use with Upsonic.

Telemetry

We use anonymous telemetry to collect usage data. We do this to focus our developments on more accurate points. You can disable it by setting the UPSONIC_TELEMETRY environment variable to false.

import os
os.environ["UPSONIC_TELEMETRY"] = "False"

🦫 PIXRA Contributors