Inferoa
An inference-native agent harness for what it calls Loop Engineering: recursive agent loops built on vLLM that account for serving, routing, context windows and prefix-cache reuse while a task runs.
Inferoa is an agent harness built around the idea that an agent loop is also an inference workload: as turns accumulate, prompt prefixes drift, cache reuse collapses, and serving and routing choices start to matter. It aims to keep the loop aware of serving, routing, context windows, prefix cache, multimodal endpoints and self-hosted model paths.
Its loop command runs durable, recursive loops that inspect, edit, test, verify, decide, remember and continue across tasks, carrying an objective and completion evidence rather than stopping after one answer. It uses independent feedback surfaces such as plans, tests, tool results and research metrics, and applies memory and context controls including compression, summaries, and bounded tool output to keep useful evidence in context without letting stale state take over.
It is built on top of the vLLM ecosystem, and its prefix-cache discipline uses prompt epochs and deterministic tool schemas to protect reusable prefixes while the loop runs.