Agentic coding actions by Fortytude
TokenDiet
Reads your repo, finds where your code wastes money on LLM calls, and opens a Pull Request that cuts the bill — with before/after proof where a model change is involved.
Credits
—
loading…
A credit is only spent when a PR with real savings opens.
Empty run? Nothing charged.
loading…
Watch = auto-mode: every push
re-scans and opens a PR when a proven saving appears. Pro includes it;
on credits, each shipped fix still costs 1 — nothing burns silently.
02 · Runs
| when | repo | status | projected $/mo | credits | |
|---|---|---|---|---|---|
| no runs yet | |||||
Click a run to see what was
found — including model changes tested and rejected by the replay-gate
(rejected = $0, never charged: that's the gate protecting you).
TokenDiet reviews the direct SDK call-sites it can detect (~80%) — honest
coverage, never "exhaustive". PRs land on a
tokendiet/vN branch;
nothing ever touches your default branch, database, or production.
We never store, log, fork or train on your code — analyzed transiently,
then discarded.