A tiny toolkit to stop AI CLI tools dying silently when one hits its usage limit
When running Codex, Claude Code, and side by side, it is common to start a long task, walk away, and return to find it failed within a minute because one provider hit its usage limit — while the others had plenty of quota sitting unused. llm-tools is a small set of three Python utilities built to fix exactly that. The tool llm-usage prints a table showing how much quota remains on each service before you start.
llm-exec runs a task and automatically skips any provider that is currently , routing the work to one that still has capacity. llm-rotate distributes tasks across providers in rotation. None of the tools call AI APIs directly; they work entirely through whichever CLI tools you already have installed, keeping setup minimal.
The project is open source under the , requires no sign-up, and sends nothing to external servers. It runs on Linux and macOS, though the suspend-and-resume feature is Linux only. Currently supported providers are Codex, Claude, and Copilot.
Key points
- llm-usage shows remaining quota for each AI service at a glance before starting work
- llm-exec automatically skips a provider and runs the task on a free one instead
- llm-rotate spreads tasks across providers in turn to avoid draining one service first
- No API keys or extra needed — it uses your already-installed CLI tools
- Open source (Apache-2.0), no sign-up, no external data sent; Linux and macOS supported