Agent-o-rama: Scalable, Traceable, Stateful AI agents in Pure Clojure or Java - Nathan Marz

Nathan Marz introduces Agent-o-rama, an open-source library for building scalable, traceable, and stateful AI agents in Clojure or Java.

The unpredictability of LLMs makes observability critical to building AI agents. You need deep tracing to diagnose issues, and you must constantly evaluate the quality of agent performance. Agent-o-rama is a new open-source library for building high-performance AI agents in Clojure or Java, built on top of Rama. Inspired by LangGraph, Agent-o-rama models agents as graphs of Clojure or Java functions that do LLM calls, interact with durable storage, or other custom logic. Agent-o-rama doesn't require learning how to program Rama and is easy to learn. Integration with local or remote LLMs, databases, and more is seamless with LangChain4j. Agent-o-rama includes a web UI for inspecting and forking agent runs, seeing stats on runtimes and token usage, and managing evaluation datasets. Unlike LangGraph and other frameworks, Agent-o-rama is free to use because it runs on your infrastructure. Traces are stored using Rama's built-in, replicated storage. So there’s no usage-based billing or sending sensitive data to third-party services. Agent-o-rama also offers better support for parallel, high-throughput workflows, and it has built-in storage that replaces the need for most databases. Whether you're experimenting with small agents or scaling up complex workflows, Agent-o-rama brings structure and visibility to the chaos of LLM-driven systems. Biography Nathan is the founder of Red Planet Labs. Previously he created a number of popular open-source projects, the most well-known being Apache Storm. Storm was the first open-source project to enable reliable processing of data at large scale in real time. Nathan has programmed primarily in Clojure since 2010. His biggest open-source contribution in Clojure is Specter, which enables data structures both simple and complex to be queried and transformed extremely concisely and with performance rivaling hand-optimized code. Nathan wrote the book "Big Data: principles and best practices of scalable realtime data systems". Video recorded Nov 14, 2025 at Clojure/Conj 2025 in Charlotte, NC. clojure-conj.org Before Red Planet Labs, Nathan was an engineer at Twitter where he founded one of their core infrastructure teams.