I built the same support ticket triage twice, with Camel's new semantic language and with LangChain4j's decision API, against a System One model running on Ollama. Here is what I learned.
Camel 4.22 introduces a unified AI tool abstraction and a built-in MCP server. Define a tool once as a Camel route, and it works with LangChain4j, Spring AI, and OpenAI.
How I built an AI agent that classifies Gmail, organizes your inbox, and drafts replies using camel-openai, structured output, and Gmail DataType Transformers.
Using camel-openai with structured output broke my local email triage agent. A system prompt pattern fixed it, and Gemma 4 took the crown.
I tested several local LLMs on my M2 Pro to build a Gmail triage agent. Here's what I learned about choosing the right model for real agentic use cases.
