A frame from the repository’s CLI demonstration using dummy-model data.
The idea
A model score can hide changes that matter to a product. insideLLMs provides a way to run shared probes across models and retain the resulting inputs, outputs and run artifacts for comparison.
The approach
The emphasis is on inspectable evidence: run a defined experiment, preserve its artifacts and compare the result with a baseline. Deterministic artifact handling makes identical recorded inputs and responses comparable; it does not make live model generation deterministic.
A shared interface for model and probe comparisons
Recorded run artifacts and behavioural diffs
An offline dummy-model path for exploring the workflow
