v0.3.0 — First Public Release

AgentEnsemble v0.3.0 — First Public Release 🎉

Production-ready multi-agent orchestration for Python. Build ReAct agents,
Swarms, Pipelines, Debate systems, Routers, Planners, and WorkflowGraphs —
with built-in observability, cost tracking, session memory, and human-in-loop
support. Comparable to LangGraph, CrewAI, and AutoGen.

---

✨ What's included

7 Agent Types

• ReActAgent — LLM-driven tool use with reasoning loops
• StateGraphAgent — custom node-based workflows
• RAGAgent — document Q&A with fallback strategies
• HybridAgent — LLM routes SEARCH → RAG → VALIDATE → ANSWER automatically
• RouterAgent — LLM picks the best agent for the task
• PlannerAgent — decomposes complex tasks into subtasks
• StructuredAgent — Pydantic-validated structured outputs

5 Orchestration Patterns

• SwarmOrchestrator — parallel agents via asyncio.gather
• PipelineOrchestrator — sequential agent workflow
• DebateOrchestrator — solvers propose → feedback rounds → aggregator votes
• Supervisor — central coordinator with optional LLM-based routing
• WorkflowGraph — composable DAG with nodes and edges

Tools

• @function_tool — decorate any function, schema auto-generated
• SearchTool — Serper or DuckDuckGo
• RAGTool — index URLs, ask questions
• ValidationTool, ToolRegistry

Production Runtime

• Runner.run() / Runner.arun() — sync and async execution
• RunConfig — retries, hooks, session, tracing
• TraceHooks — full observability: run/LLM/tool events, token usage, cost estimation
• InMemorySession / SQLiteSession — multi-turn context memory
• interrupt_before_tools — pause before sensitive tools, resume with human approval

---

⚡ Quick Start

pip install agentensemble==0.3.0
# or simply
pip install agentensemble

from agentensemble import ReActAgent, SearchTool, Runner

agent = ReActAgent(name="research", tools=[SearchTool()], max_iterations=5)
result = Runner.run(agent, "What are the latest AI breakthroughs in 2024?")
print(result["result"])

---

🔍 Observability out of the box

from agentensemble import ReActAgent, Runner, RunConfig, TraceHooks, SearchTool

agent = ReActAgent(name="research", tools=[SearchTool()], max_iterations=3)
config = RunConfig(trace_hooks=TraceHooks())
result = Runner.run(agent, "Who won the 2024 Nobel Prize in Physics?", config=config)

print(result["metadata"]["total_usage"])  # {'input_tokens': 1247, 'output_tokens': 89, ...}
print(result["metadata"]["cost_usd"])     # 0.00293

---

🤝 Human-in-loop

config = RunConfig(interrupt_before_tools=["execute_payment"])
result = Runner.run(agent, "Pay $50 to Alice", config=config)

if result.get("__interrupted__"):
    print("Paused for approval:", result["pending_tool"])
    result = Runner.run(agent, "...", config=config,
                        resume=result["run_state"], resume_value="Approved")

---

📊 How it compares

Feature	AgentEnsemble	LangGraph	CrewAI	AutoGen
Async API	arun() everywhere	Yes	Yes	Yes
Debate pattern	✅ DebateOrchestrator	Manual	Yes	Yes
Graph workflows	✅ WorkflowGraph + StateGraph	Native	No	No
Session memory	✅ InMemory + SQLite	Checkpointer	No	No
Observability	✅ TraceHooks, usage, cost	LangSmith	No	No
Human-in-loop	✅ interrupt_before_tools	interrupt()	No	No

---

🔗 Links

• PyPI: https://pypi.org/project/agentensemble/
• Showcase: examples/showcase_all.py
• Docs / Roadmap: docs/ROADMAP_2025.md
• Author: Irfan Ali — DataCortex IQ