H2O Agents: Building Tool-Using Workflows with h2oGPTe
- Agentic AI
- Enterprise h2oGPTe
- H2O Driverless AI
This lesson walks through the building blocks behind AI agents, starting with LLM chains as fixed pipelines and then agents that use reasoning and planning to choose which tool to call next. It introduces the GAIA benchmark, a set of human-designed questions used to measure an assistant's reasoning, multimodal, and tool-use abilities. The lesson then demonstrates Enterprise h2oGPTe's agent settings, including accuracy level and available tools, before walking through examples: a weather-based question, a PDF report with citations, plotting stock prices in Python, and calling H2O Driverless AI to build a churn model.
Key takeaways
- How agents differ from LLM chains by using reasoning and planning to decide which tool to call next
- Why the GAIA benchmark tests real-world reasoning, multimodal handling, and tool use rather than recalled facts
- How to configure an h2oGPTe agent's accuracy level and maximum turns to trade off speed against depth
- How h2oGPTe agents can call H2O Driverless AI directly to train and evaluate a churn prediction model
- The option to have an agent generate downloadable Python code, PDFs, and images alongside its answers
Start owning your AI