Free spend scan
See and govern your AI spend, from your own logs, not a calculator.
Send a sample of your request logs. Get a gap report in 48 hours: where your spend goes, what’s controllable today, and what could be saved (concierge-verified or proof-pending). Free. No proxy. No commitment.
Log export
Metadata: model, tokens, timestamps
Read-only access
A few hours/days of traffic
Config paste
Routing config + prompt set
RECOVEA · SPEND SCANSample data~$184K/mo · 142K reqs / 3d
| lever | risk | recoverable |
|---|---|---|
| Prompt-cache prefix hygiene (planned) | Zero | measured per route |
| Batch API migration (planned) | Zero | provider Batch tier |
| Reasoning / effort trimming (planned) | Low | measured per route |
| Dedup / single-flight (concierge) | Zero | concierge-verified, not self-serve billed |
| Exact cache (concierge) | Zero | concierge-verified, not self-serve billed |
| Model routing (eval-gated, shadow) | Med | on the routable slice, gated |
| Context compression / RAG trim (planned) | Med | input trim on RAG route |
control live today · verified savings off · proof pendingyour spend, attributed
Illustrative; results vary. Cost visibility and control are live today. Cache/dedup savings are concierge-verified, never a self-serve billed dollar; eval-gated savings land as the proof gate ships — measured from your own traffic, net of quality. They estimate; we settle.
It’s your data, not a brochure
- Derived from your real logs, not a generic calculator.
- Honest about uncertainty: ranges, “net of quality, to be measured from your own traffic.”
- Maps directly to the levers we’d actually run.
Privacy
Metadata-only is fine. No prompt bodies retained by default. We tell you what we keep and for how long, and delete on request. Read our security posture.