A transparent estimate. It counts the tokens memory saves (avoided re-derivation and repeated mistakes) and the tokens the injected block costs — so the number is honest. Move the sliders to match your setup.
Your setup
20
1
3 avoided
Facts the memory block carries, so the agent skips re-reading these files each session.
1,500
1
One wrong-turn cycle ≈ wrong output + your correction + regeneration.
900
Your time — the real value
10
Re-explaining context, waiting on re-derivation, fixing the same mistake again.
The cost side
400 tokens
40
Estimated net savings
—
tokens / month
Your time — the real value
—
— hours / month
Per session — where it comes from
Cold-start re-read
Mistakes avoided
− Injection cost
− Capture (amp_write)
Net / session
Read this before quoting a number. This is a model, not a measurement — the real figure depends on your codebase and workflow. The biggest win is your time and lost focus — which is why the dollar figure here is based on time saved, not tokens (token cost is negligible and varies by plan anyway). With caching off, a long session's per-turn block cost can eat into savings that are concentrated at session start — which is exactly why infernoflow keeps the injected block lean (~4 entries). infernoflow has no built-in savings meter today; these are estimates from the assumptions above.