Gio 🌱 — token savings

Repo click · 1,158 chunks · 75 files · 8 probe questions

13.6×
fewer input tokens than reading the whole files an answer lives in
274,530
input tokens saved (this run)
$0.82
saved at sonnet input pricing
13.6×
more questions in the same budget

Per question — tokens read

Gio (targeted spans) Reading whole files
where is the main entry point
3,262
47,371
how is configuration loaded
1,513
21,230
where are errors handled
1,732
43,047
how is authentication or access control handled
3,403
45,668
where is input data validated
3,066
20,199
how are results formatted or rendered
2,781
47,125
where is logging set up
2,500
27,877
how are external services or APIs called
3,601
43,871

Gio fed 21,858 tokens total vs 296,388 to read whole files. Deterministic, local, char/4 token estimate — no API key. Retrieval quality (hit@k) is a separate labeled eval (scripts/eval_retrieval.py).