Upsonic

Agent Framework For Fintech and Banks

7.8k
Stars
+60
Stars/month
10
Releases (6m)

Star Growth

+8 (0.1%)
7.7k7.8k8.0kMar 27Apr 1

Overview

Upsonic is an open-source AI agent framework designed for building production-ready agents with a focus on safety and compliance. The framework supports multiple AI providers including OpenAI, Anthropic, Azure, and Bedrock, making it provider-agnostic for enterprise deployments. It features built-in safety policies, OCR capabilities, persistent memory, session management, and multi-agent coordination. Originally positioned for fintech and banking use cases, Upsonic emphasizes compliance monitoring and includes tools for financial analysis, document processing, and automated research workflows. The framework provides a Python-based approach to agent development with MCP tool integration and support for complex multi-agent orchestration. With over 7,800 GitHub stars, it represents a growing solution for organizations requiring AI agents with enterprise-grade safety controls and multi-provider flexibility.

Deep Analysis

Key Differentiator

AI agent framework with built-in safety engine (PII anonymization, content policies) and OCR — vs frameworks like CrewAI or AutoGen that lack native safety controls

Capabilities

  • Production-ready AI agent framework with safety engine
  • Autonomous agent with filesystem and shell access (sandboxed)
  • OCR support with multiple engines (EasyOCR, PaddleOCR, Tesseract)
  • Policy-based content filtering (PII anonymization, toxicity)
  • Memory management with session and long-term storage
  • Multi-agent teams (sequential and parallel coordination)
  • MCP tool integration and human-in-the-loop workflows

🔗 Integrations

OpenAIAnthropicAzureAWS BedrockMCP protocolEasyOCRPaddleOCRTesseractKubernetes

Best For

  • Teams needing AI agents with built-in safety policies (PII, compliance)
  • Document processing workflows with OCR + AI agents
  • Production deployment of sandboxed autonomous agents

Not Ideal For

  • Simple chatbot applications without safety requirements
  • Teams wanting a no-code agent builder

Languages

Python

Deployment

pip packageAgentOS (Kubernetes-based)Self-hosted

Pricing Detail

Free: Open source MIT, full framework free
Paid: N/A — open source, but requires LLM API costs

Known Limitations

  • Rebranded from GCA to Upsonic, documentation may reference old name
  • AgentOS deployment requires Kubernetes
  • OCR requires additional package install
  • Relatively smaller community compared to LangChain/CrewAI

Pros

  • + Multi-provider AI support (OpenAI, Anthropic, Azure, Bedrock) with unified interface
  • + Built-in safety policies and compliance monitoring for enterprise environments
  • + Comprehensive agent capabilities including memory, OCR, and multi-agent coordination

Cons

  • - Python-only implementation limits cross-language integration
  • - Smaller community compared to major AI frameworks
  • - Documentation hosted externally rather than in-repository

Use Cases

  • Financial analysis and reporting with automated data processing and insights generation
  • Document analysis and processing using OCR to extract text from images and PDFs
  • Multi-agent workflow orchestration for complex research and data gathering tasks

Getting Started

Install via pip with 'uv pip install upsonic', create an Agent instance specifying your preferred AI model (e.g., 'anthropic/claude-sonnet-4-5'), then define and execute Task objects with your agent using the print_do() method

Compare Upsonic