Jeremiah
cf.run()
Start with a one-liner, then gradually take control of the entire agentic loop, as needed: summary = cf.run("Summarize this article", context=dict(article=article))
🧠 Stateful Flows + History
Every action is tracked automatically, so you can maintain context across multiple agent turns, even if you run custom Python code in between.
⚙️ Rewritten Orchestration Engine
A new, event-based orchestrator assigns agents to work and dynamically compiles prompts for each agent's LLM model, objectives, and workflow visibility.
🤝 Agent Collaboration Strategies
Control how AI agents take turns when collaborating on tasks: round-robin, random, moderated, delegation, and more.
💬 Enhanced User Interaction
ControlFlow is code-first, chat-second, but you can put a human-in-the-loop with a keyword: cf.run("What's up, doc?", interactive=True)
✅ Structured Output & Validation
Ensure that AI results meet your requirements with structured outputs and custom validation.
🔗 Full Prefect 3.0 Integration
Every ControlFlow flow is a Prefect flow, for total visibility and additional control. Mix traditional and AI tasks seamlessly.
📚 All-New Docs & Examples
Learn faster, build better at https://controlflow.ai
Upgrade now: pip install -U controlflow
Happy engineering! 🚀🧠🤖