Priced per engineer covered, not per employee. We make the same whether you route $1 or $1M of inference.
Building AI into your product? Meter prices per request →
Connect one source of spend and get a one-page AI cost report on your own data. Enough to know whether the waste is worth chasing. No card, no commitment.
The full read-only audit, running continuously. Every surface, every connector, and the monthly P&L your finance team signs. It reads and recommends. It never touches your systems.
Start Observe — $16/engineer→Everything in Observe, plus opt-in guardrails that act on the findings. Cap runaway spend. Route work to a cheaper model. Both run on your own provider keys, flat per engineer either way.
Request accessor book a 30-min call →Observe, worked out: 50 engineers = $800/mo · 200 engineers = $3,200/mo
If your first Snapshot finds less recoverable spend than a year of Observe costs, don't buy it. If you already have, tell us within 30 days and we refund what you paid.
That is a contractual commitment, not a marketing line — §4 of the terms carries it word for word.
Choose month-to-month, or an annual (1-year) term for 10% off.
Regulated or multi-region? Add SSO and SCIM, role-based access control, a single-tenant region, a SOC 2 readiness report (20 of 34 controls met as of 26 July 2026, with 9 partial and 5 deferred) with security-review support, and a DPA. Available on either paid edition. Same platform, plus the controls procurement asks about on the first call.
Billing follows the engineers the audit covers. Non-technical staff are never counted.
Observe pays for itself above roughly $150 per engineer per month of seat and subscription spend. Below that, use the free Snapshot — it's free for a reason.
Control routes through your own vendor keys, so inference is billed to you directly.
Proof levels, visible formulas, and read-only access are not upsells. Observe reads and recommends. Control adds the ability to act, on your keys and on your terms.
| Capability | Snapshot | Observe | Control |
|---|---|---|---|
| AI cost report (one-pager) | |||
| Cross-vendor spend normalization | |||
| Evidence tiers on every number | |||
| Read-only connectors | snapshot only | all | all |
| All product surfaces | – | ||
| Continuous findings, recomputed daily | – | ||
| Monthly P&L, editable assumptions | – | ||
| Team-level productivity and what the speed cost you | – | ||
| Routing and throttling recommendations | – | ||
| Continuous monitoring: utilization, spend, usage, output | – | ||
| Drift and anomaly alerts, weekly digest | – | ||
| Control acts on the findings (opt-in) | |||
| Spend throttles and budget guardrails | – | – | |
| Model routing on your own keys | – | – | |
| Shadow mode and measured route savings | – | – | |
| Bring-your-own-keys, zero token margin | – | – | |
| Enterprise controls, annual add-on for either paid plan | |||
| SSO and SCIM provisioning | – | add-on | add-on |
| Role-based access control | – | add-on | add-on |
| Single-tenant region of your choice | – | add-on | add-on |
| SOC 2 readiness report and DPA | – | add-on | add-on |
Observe is the read-only audit. It reads every source, finds the waste, and recommends where spend should move. It never touches your systems. Control adds the ability to act on those findings. It throttles runaway spend and routes work to a cheaper model, using your own provider keys. Most teams start on Observe and add Control once they trust the numbers.
Snapshot is a one-time free look. Observe runs continuously and re-audits every day, so you always see current numbers. It also watches for drift, alerts you, and sends a weekly digest, all included. That is what keeps the recovered spend from creeping back.
No. Control is flat per engineer and routes through your own vendor keys, so the inference bill stays entirely yours. That is what keeps the audit independent.
Observe is read-only by architecture and cannot write a byte back. Control is the one place Spendassay can act. It only acts on guardrails you set, it runs on your own keys, and it never reads your code or prompts. You can run Control in shadow mode first. Shadow mode scores a routing policy against real traffic without changing anything.
Anyone whose AI seats or usage the audit ingests. You set the roster and billing follows it. Finance, sales, and everyone else on ChatGPT do not count against your bill.
No. Start free with a Snapshot on one vendor, or a CSV export from billing. Every connector has a CSV or export fallback, so you can run a full Snapshot from exports alone — no admin access, no OAuth grant, nothing for IT to approve. That also covers the case where your plan tier has no API at all.
Yes, in §4 of the terms, in those words. If you already have, tell us within 30 days and we refund what you paid. It is a refund of the term you paid for, claimable within 30 days of that first Snapshot. Two things it is not: it is not a promise that we will find a particular amount (§8 says so explicitly), and it is not a trial — Snapshot is already free, so you can see the numbers before you pay us anything at all.
Run the audit on your own stack. The first Snapshot is free and lands the same day. No card, and no connectors needed to look around first.