Q: Will SpendTensor add latency to my requests?

Our edge proxy adds <3ms at p99. We run on 240+ PoPs globally and use HTTP/3 with connection reuse to your upstream providers, often making requests faster than going direct.

Q: How does the routing engine maintain quality?

Every recommended model swap is benchmarked against your real traffic with held-out evals. We only route when output quality meets the threshold you set (default 0.95 similarity).

Q: What happens if SpendTensor is unreachable?

The proxy fails open. If our edge cannot serve a request it is passed straight through to your upstream provider, so an incident on our side never becomes an outage on yours.

Q: Does caching risk returning stale answers?

Semantic caching is opt-in per route with a configurable similarity threshold and TTL. You can exclude endpoints, users, or prompt patterns entirely.

Q: Do you store prompts or completions?

By default we store only metadata — model, tokens, latency, fingerprints. Full payloads are opt-in per workspace and can be redacted, sampled, or stored in your own bucket.

Q: How are our provider API keys stored?

Keys are encrypted with AES-256-GCM before they touch the database and are only decrypted inside the sync worker at request time. They are never returned to the browser.

Question 1

How fast can we get up and running?

Accepted Answer

Most teams are live in under 10 minutes. Point your existing OpenAI / Anthropic SDK at api.spendtensor.io — done. We auto-detect every model and start tracing immediately.

Question 2

Do we have to change our application code?

Accepted Answer

Only if you want real-time tracing. Proxy mode is a one-line base URL change; read-only key mode requires no code change at all — we pull usage from each provider's billing and usage APIs on an hourly schedule.

Question 3

Can you backfill our historical spend?

Accepted Answer

Yes. On connection we backfill up to 90 days of provider usage so your first dashboard shows a real baseline and trend rather than an empty chart.

Question 4

Which providers do you support?

Accepted Answer

OpenAI, Anthropic, Google Gemini, Azure OpenAI, and AWS Bedrock today, plus OpenAI-compatible self-hosted endpoints such as vLLM. Costs are normalized across all of them.

Question 5

Will SpendTensor add latency to my requests?

Accepted Answer

Our edge proxy adds <3ms at p99. We run on 240+ PoPs globally and use HTTP/3 with connection reuse to your upstream providers, often making requests faster than going direct.

Question 6

How does the routing engine maintain quality?

Accepted Answer

Every recommended model swap is benchmarked against your real traffic with held-out evals. We only route when output quality meets the threshold you set (default 0.95 similarity).

Question 7

What happens if SpendTensor is unreachable?

Accepted Answer

The proxy fails open. If our edge cannot serve a request it is passed straight through to your upstream provider, so an incident on our side never becomes an outage on yours.

Question 8

Does caching risk returning stale answers?

Accepted Answer

Semantic caching is opt-in per route with a configurable similarity threshold and TTL. You can exclude endpoints, users, or prompt patterns entirely.

Question 9

Do you store prompts or completions?

Accepted Answer

By default we store only metadata — model, tokens, latency, fingerprints. Full payloads are opt-in per workspace and can be redacted, sampled, or stored in your own bucket.

Question 10

How are our provider API keys stored?

Accepted Answer

Keys are encrypted with AES-256-GCM before they touch the database and are only decrypted inside the sync worker at request time. They are never returned to the browser.

Question 11

Which compliance frameworks do you cover?

Accepted Answer

SOC 2 Type II, GDPR with EU data residency, ISO 27001 aligned controls, and HIPAA with a BAA for eligible plans. Annual third-party penetration tests are shared on request.

Question 12

Can we restrict who sees cost data?

Accepted Answer

Yes. Role-based access control is available from the Business plan and scopes dashboards, budgets, and exports by workspace and role.

Question 13

How are you priced?

Accepted Answer

Self-serve plans are flat monthly. Enterprise is a percentage of verified savings — typically 10–15% — so we only win when you do.

Question 14

Do you charge per seat?

Accepted Answer

No. Every plan includes unlimited teammates — pricing is based on capability tier, not headcount.

Question 15

Can we change or cancel plans?

Accepted Answer

Upgrade, downgrade, or cancel at any time from the dashboard. Changes prorate on your next invoice and your historical data stays intact.

Question 16

Do you offer annual invoicing?

Accepted Answer

Yes — annual invoicing and purchase orders are available on Business, Scale, and Enterprise agreements.

Question 17

Can we self-host?

Accepted Answer

Yes. Enterprise customers can deploy SpendTensor entirely in their own VPC (AWS, GCP, Azure). The control plane is air-gapped and we ship monthly versioned releases.

Question 18

Do you support SSO and SCIM?

Accepted Answer

SAML SSO is included from the Scale plan; SCIM user provisioning is available on Enterprise agreements.

Question 19

What does onboarding look like?

Accepted Answer

Enterprise customers get a named account manager, a guided rollout plan across services, and a verified savings review at the end of the first quarter.

Question 20

Can you sign a custom DPA or MSA?

Accepted Answer

Yes. Our legal team works from your paper for Enterprise agreements, including regional data-processing addenda.

Questions, answered.

Getting started

Performance & routing

Security & data

Pricing & billing

Enterprise

Need a security review packet?

Still have questions?