sota-agent-framework 0.4.2

Domain-agnostic agent framework for integrating AI agents into data pipelines

Agent Framework

Production-ready template for building AI agent workflows in any domain.

Build intelligent agents with memory, reasoning, optimization, and seamless Databricks integration. Start simple, scale to autonomous systems.

---

🚀 Quick Start

Installation

# Basic (core features only)
pip install agent-framework

# With features you need
pip install agent-framework[all]  # Everything
pip install agent-framework[databricks]  # Databricks integration
pip install agent-framework[optimization]  # DSPy + TextGrad

Choose Your Path

🤖 Have a Use Case? (NEW! - AI-Powered)

# From text
agent-architect "Build a fraud detection system with memory and self-improvement"

# From document (txt, md, pdf, docx)
agent-architect --file requirements.txt
# → Instant architecture recommendation: Level, schemas, features, integrations!

Describe your use case in natural language or provide a document, get instant recommendations.

🎓 Want to Learn? (NEW!)

agent-learn  # Interactive learning mode - build 5 progressively complex examples

Learn by building: chatbot → context-aware → production API → complex workflow → autonomous multi-agent

🚀 New to Agents?

agent-setup  # Interactive wizard guides you through

🔧 Building an Agent?

agent-generate --domain "fraud_detection" --output ./my-agent
cd my-agent && agent-advisor .  # Get recommendations

⚡ Expert?

# Use the framework as a library
from agents import Agent, AgentRouter
from memory import MemoryManager
from orchestration import AgentWorkflowGraph

📖 See complete getting started guide →
🎓 See learning path →

---

✨ Key Features

Core Framework

• ⚡ Multiple Execution Modes - In-process, parallel, Ray, serverless
• 🔌 Pluggable Architecture - Use only what you need
• 📝 Type-Safe Schemas - Pydantic models throughout
• ⚙️ YAML Configuration - Infrastructure as code

Agent Intelligence

• 🧠 Agent-Governed Memory - Smart storage, retrieval, reflection, forgetting
• 🎯 Reasoning Optimization - Trajectory tuning, CoT distillation, self-improvement
• 🔄 Plan-Act-Critique Loops - LangGraph-powered orchestration
• 🤝 A2A Protocol (Official) - Linux Foundation standard for cross-framework agent communication
• 📊 Comprehensive Benchmarking - 6+ metrics, regression testing

Production Ready

• 🏢 Databricks Native - Unity Catalog, Delta Lake, MLflow integration
• 📈 Complete Observability - OpenTelemetry, execution graphs, trace replay
• 🔧 Prompt Optimization - DSPy & TextGrad for auto-tuning
• 🌐 REST & WebSocket APIs - Production services included
• 🎛️ Experiment Tracking - Feature flags, A/B testing, MLflow

Developer Experience

• 🎯 Progressive Disclosure - Strong defaults for beginners, full control for experts
• 🤖 AI-Powered Tools - agent-architect, agent-setup, agent-generate, agent-advisor, agent-benchmark, agent-learn, agent-deploy
• 📚 8 Core Docs - Clear, concise, use-case driven
• 🔍 Use-Case Guidance - Know exactly which features you need
• 🚀 Deployment Ready - Docker, K8s, Databricks, Serverless templates included

---

🏗️ Technology Stack

Built on industry-leading technologies for production-grade AI agents:

Component	Technology	Production Config
Agent Runtime	Databricks Apps (hot pools)	min_instances: 2, scale_to_zero: false
LLM Inference	Databricks Model Serving	Always-on (no scale-to-zero)
Orchestration	LangGraph + Databricks Workflows	Plan → Act → Critique loops
Agent Memory	Lakebase + Delta Lake (UC)	Async vector + metadata queries
A2A Transport	FastAPI/Starlette (in container)	JSON-RPC 2.0, peer-to-peer
MCP Servers	FastAPI/Starlette (in container)	Tool/resource discovery
Telemetry	OTEL → ZeroBus → Delta Lake	Batch writes (10s/1000 events)
Prompt Registry	Unity Catalog Volumes	Version-controlled, auto-refresh
Prompt Optimization	DSPy + TextGrad (offline jobs)	Scheduled (nightly), no runtime overhead
Tracing & Evaluation	Databricks MLflow	Experiment tracking, model registry
Dashboards	Databricks SQL	Real-time agent metrics

---

📦 Use Cases

Works for any agent workflow:

• 🔒 Fraud Detection & Risk Analysis
• 💬 Customer Support & Chatbots
• 📝 Content Moderation
• 🏥 Healthcare & Diagnostics
• 🔍 Data Quality & Anomaly Detection
• 📊 Analytics & Report Generation
• 🤖 Your Use Case Here

---

📖 Documentation

Start Here:

1. Getting Started - 5-minute setup
2. User Journey - Choose your path (Beginner/Intermediate/Advanced)
3. Feature Selection - Which features do YOU need?

Core Guides:

• Architecture Advisor - AI-powered recommendations
• Learning Path - Learn by building (5 levels)
• Integrations - MCP, LangGraph, Databricks, A2A
• Advanced Features - Memory, Reasoning, Optimization, Benchmarking

Quick Links:

• Documentation Map - Complete navigation guide
• Examples - Working code examples
• Benchmarks - Evaluation suites

---

🛠️ CLI Tools

# 🎓 Interactive learning mode (NEW!)
agent-learn              # Learn by building 5 progressively complex examples
agent-learn start 1      # Start Level 1: Simple Chatbot
agent-learn start 2      # Start Level 2: Context-Aware Assistant

# Interactive setup wizard (use-case based)
agent-setup

# Generate new project
agent-generate --domain "your_domain" --output ./project

# Analyze project & get recommendations
agent-advisor ./project

# Run benchmarks & evaluations
agent-benchmark run --suite fraud_detection --report md

# Deploy to production (NEW!)
agent-deploy init --platform kubernetes  # Generate deployment configs
agent-deploy build --tag v1.0.0          # Build Docker image
agent-deploy status                       # Check deployment readiness

---

🎯 Feature Selection Guide

Use Case	Memory	Reasoning	Optimization	Monitoring	LangGraph
Simple Chatbot	⚪ Optional	❌ No	❌ No	⚪ Optional	❌ No
Context-Aware Agent	✅ Yes	⚪ Optional	⚪ Optional	✅ Yes	⚪ Optional
Production API	⚪ Optional	❌ No	⚪ Optional	✅ Yes	❌ No
Complex Workflows	✅ Yes	✅ Yes	⚪ Optional	✅ Yes	✅ Yes
Autonomous Agent	✅ Yes	✅ Yes	✅ Yes	✅ Yes	✅ Yes

📖 See detailed feature guide →

---

🏗️ Architecture

Agent Framework
├── agents/           # Core agent classes & registry
├── memory/           # Agent-governed memory system
├── reasoning/        # Trajectory optimization & feedback
├── optimization/     # DSPy & TextGrad prompt optimization
├── orchestration/    # LangGraph workflows
├── evaluation/       # Benchmarking & metrics
├── visualization/    # Databricks-native observability
├── telemetry/        # OpenTelemetry → Delta Lake
├── uc_registry/      # Unity Catalog integration
├── experiments/      # Feature flags & A/B testing
├── monitoring/       # Health checks & metrics
├── services/         # REST API & WebSocket
└── infra/            # Terraform for Databricks

📖 See detailed architecture →

---

🚀 Example: Fraud Detection Agent

from agents import Agent, CriticalPathAgent
from memory import MemoryManager
from orchestration import AgentWorkflowGraph

# Define agent
class FraudDetectorAgent(CriticalPathAgent):
    async def process(self, input_data):
        # Check memory for similar cases
        similar = await self.memory.retrieve(
            query=f"transaction {input_data.transaction_id}",
            top_k=5
        )
        
        # Run detection
        result = await self.detect_fraud(input_data)
        
        # Store in memory
        await self.memory.store(result, importance="HIGH")
        
        return result

# Use with LangGraph for complex workflows
workflow = AgentWorkflowGraph(agent_router=router)
workflow.add_node("planner", PlannerNode())
workflow.add_node("detector", FraudDetectorAgent())
workflow.add_node("critic", CriticNode())

result = await workflow.run(transaction_data)

📖 See more examples →

---

🤝 Contributing

We welcome contributions! See our contribution guidelines (coming soon) or file an issue.

---

📄 License

MIT License - see LICENSE for details.

---

🔗 Links

• PyPI: https://pypi.org/project/agent-framework/
• GitHub: https://github.com/somasekar278/universal-agent-template
• Documentation: DOCUMENTATION_MAP.md

---

⭐ What Makes This Agent?

Unlike orchestration-only or research-only agent frameworks, SOTA Agent ships a complete agentic development stack including autonomous planning loops, agent-governed memory, reasoning trajectory optimization, prompt auto-tuning, benchmark harnesses, and governed deployment — built for real data pipelines and production SLAs

✅ Agent-Governed Memory - Not just storage, intelligent decisions
✅ Plan-Act-Critique Loops - True autonomous workflows
✅ Reasoning Optimization - Learn from execution trajectories
✅ Comprehensive Benchmarking - Track performance over time
✅ Databricks Native - Production-ready from day one
✅ Progressive Disclosure - Works for beginners AND experts
✅ Modular Design - Use only what you need

🚀 Get started now →