What is inferoa?

Inferoa is a command-line agent harness built around the idea of running AI in recursive loops until a task is provably complete. Rather than asking an AI for one answer at a time, the loop approach gives it an objective, tools to inspect and modify code or files, ways to verify its own results, and the ability to keep going through multiple attempts until the verification passes. The harness manages the state of that loop, including which tasks are done, what evidence exists, and what decisions were made. The project puts particular emphasis on what it calls tokenmaxxing and inference-awareness. As an AI loop runs over many turns, the conversation history grows and earlier prefixes in the prompt may shift, which breaks the server-side cache that normally makes repeated calls faster and cheaper. Inferoa tracks this and structures each turn to preserve the reusable parts of the prompt, bound how much history accumulates, and compress or summarize stale context before it crowds out useful information. Routing is also exposed: the tool can direct different turns to different models based on cost, privacy requirements, capability needs, or whether a self-hosted model is sufficient. It integrates with vLLM, an open-source framework for high-throughput AI serving, for the actual model calls. The terminal interface opens with the inferoa command after a one-time setup that configures the model endpoint and API credentials. The /loop slash command starts a long-horizon loop from a plain-language objective. The /plan command turns a vague task into an inspectable plan before execution starts. The /tokenmaxxing command displays a live summary of token usage, prefix-cache reuse, context savings, and model-selection pressure for the current session. It is a developer-facing tool intended for engineering and research tasks that require repeated iteration, inspection, editing, testing, and verification rather than a single-shot answer.