GitHub radar
AxisAgentic: Open Runtime for Long-Horizon AI Agents
XYZ AI Lab open-sourced AxisAgentic, a Python runtime for AI agents that tackle long multi-step tasks while recording every action for replay, evaluation, and model fine-tuning.
AxisAgentic is an open-source Python runtime from XYZ AI Lab for running AI agents on long, multi-step tasks. Every agent action is written to an append-only trace log, which can replay exactly what the model saw at any stage, evaluate performance across benchmarks like BrowseComp and Humanity's Last Exam, or export curated trajectories for supervised fine-tuning. The flagship example built on top of it is XYZ-Aquila, a deep web search agent that topped common baselines on six agentic search benchmarks. The framework works with any OpenAI-compatible model endpoint.
Why a vibe-coder should care
For anyone building AI agents, AxisAgentic solves two hard problems at once: it keeps an agent on track across a long run with context management and automatic recovery, and it turns every run into training data without extra labeling — the traces collect themselves. If you want an AI that searches and reasons for minutes rather than seconds, this is the open infrastructure to start from.
▌ More finds