Built a Python agent framework where a decision model chooses actions instead of an LLM - looking for feedback
retained text
I've been building DecisionLoop, a Python framework for agents where a small decision model handles intent detection, tool selection, arguments, and control flow instead of using an LLM for every decision. I tested it on three use cases: order support, a flight desk, and a banking desk. My early tests showed better tool-call accuracy and lower cost and latency than a LangChain LLM agent using a 12B model, although results varied with smaller models. These are preliminary results, not a proper benchmark yet. I'd love feedback on the architecture, real-world use cases, difficult conversations to test, and how to benchmark it fairly against stronger LLMs. I'm also interested in building a multi-agent architecture on top of DecisionLoop, with a router delegating requests to specialist agents.
document fetched 2026-10-14 00:19 · canonical https://reddit.com/r/AI_Agents/comments/1x4os9g/built_a_python_agent_framework_where_a_decision