Last released Aug 25, 2026
Fail-safe client discipline for local LLMs on memory-constrained machines: preflight memory checks, starvation-aware retries, thinking-mode-off, token budgets.