Offline compilation builds using cached generated bodies without contacting an agent. Enable it with -fllm-offline when the implementations needed by the source have already been generated and reviewed.

$ llmc++     -fllm-offline     main.cpp     -o main

A usable cache hit supplies its body to the normal rewrite and compilation workflow. A missing or expired entry causes an error instead of silently starting a model exchange. Compiler context and cache identity must still match, unless an explicit key selects the implementation.

A function-level offline setting can require this policy for one target. It cannot be combined with function-level no_cache. Forced regeneration takes precedence over offline mode, so combining the two permits generation rather than enforcing a model-free build.

Keeping reviewed cache entries with a project makes repeat builds possible without model credentials. This controls reuse of generated bodies; it does not freeze compiler versions, libraries, or other build inputs. Behavioral tests remain useful when those inputs change.