Savings
Spend JamJet actually recovered by enforcing your policies. Metered from real behavior, not self-reported.
Savings
The Savings surface (dashboard route /dashboard/savings, under the Enforce chapter) shows the spend JamJet actually recovered by enforcing your policies. Every dollar here is measured from real enforced behavior, not a recommendation someone marked as done. That is the whole framing: metered, not self-reported. If the number went up, it is because enforcement ran and changed what your agents spent.
The headline figure is total recovered over the last 30 days. Below it, a stacked-by-source area chart plots recovered spend over time, and a "by source" bar breakdown shows how much each source contributed.
Sources
Recovered spend is attributed to the policy mechanism that produced it. Two sources contribute today.
| Source | Driven by | How it's valued |
|---|---|---|
| Cache inject | cache_inject policies | Cache-read tokens times the per-million-token price difference between full input and cache-read input, using a versioned price table. |
| Tool compaction | The tool_compaction policy | Each event's recorded tokens_saved times the input price per million tokens. |
Both valuations price the saved tokens against a versioned price table, so the recovered figure tracks the real model pricing in effect rather than a flat estimate.
Where the data comes from
The surface reads from GET /v1/cost-findings/recovered. The response carries three shapes:
- A total series, which drives the 30-day headline.
- A by-kind split, which drives the "by source" bar breakdown.
- A per-day-per-kind series, which drives the stacked area chart.
Savings is the metered counterpart to Optimize. Optimize is where you turn a finding into an enforced policy. Savings is where you see what that enforcement recovered.