Highlights from an evening on agent runtimes, model deployment, inference orchestration, memory, and open-source post-training.
Inside Open Infra Night: Building the Infrastructure Behind AI Agents
On October 22, Strukto.AI, Infron AI, and XTrace hosted Open Infra Night for Agent Builders in San Francisco. Engineers, founders, and open-source contributors came together to compare the infrastructure required to move AI agents and open models from prototypes into production.
The evening followed the production stack from agent environments and model deployment to inference, memory, and post-training. The common thread was practical: how to make these systems easier to run, adapt, and improve as real workloads evolve.
Strukto.AI covered the agent environment layer. Zecheng Zhang introduced Mirage and AgentVFS, spanning the virtual terminals, filesystems, command-line interfaces, and runtimes agents need to work with tools in a structured environment.
Infron AI focused on inference orchestration. Lawrence Xu shared how teams can move between serverless APIs, elastic capacity, hourly GPUs, and committed deployments as traffic, control requirements, and unit economics change.
XTrace addressed memory and feedback at the agent harness layer. Felix Meng discussed how useful lessons can carry across sessions, helping teams enforce engineering standards, reduce repeated corrections, and identify wasted token spend.
Thank you to Zecheng Zhang, Yue Xu, Vanessa Lu, Yifei Fang, Felix Meng, Vishvesh Bhat, Raj Rao, Yiwen Yuan, Yiming Zhang, Christine (Xinyu) H., He Qu and Fuyao(Felix) Li for contributing talks, lightning talks, and live demos. Together, their sessions expanded the conversation from optimized model deployment to open-source LLM post-training, model inference, AI-native harnesses, visual agents, and other production tooling.
Thank you as well to every attendee, to Persona for the venue, and to our co-hosts Strukto.AI and XTrace for helping bring the program together.
Open Infra Night was part of Open Source AI Week, stewarded by the Open Intelligence Foundation. We were glad to contribute a builder-led conversation to a week dedicated to open-source AI, and to continue the discussion around the infrastructure that makes it production-ready.
