Our answers to the LLM gateway security questionnaire
Standard Compute's completed answers to the LLM gateway security questionnaire, ready to paste into your vendor review. Each answer is one or two sentences and links to the page that backs it: our security overview, sub-processor list and DPA, which you sign in the dashboard.
Last updated
Download as CSV
Create a company workspaceor send us your own questionnaire at contact@standardcompute.com
Data residency and regional processing
| ID | Question | Answer | Evidence |
|---|---|---|---|
| DR-01 | Where are prompts and completions processed (inference), and can processing be restricted to a named region such as the EU/EEA? | Set a Europe rule for your company, a team or a single key, and every request, including retries and fallbacks, is served only by approved European inference endpoints. Our EU gateway terminates TLS and runs in Finland, and if no European endpoint can serve a request, it is refused before your prompt is sent. | SecurityEuropean gateway |
| DR-02 | How does the vendor define the region? For example, does "Europe" mean EU/EEA only, or does it also include the UK, Switzerland or other countries? | A Europe rule admits only the approved European inference endpoints named on our sub-processor list, each with its processing location. | Sub-processors |
| DR-03 | Do retries, fallbacks, classifiers, guardrails, caching and logging stay inside the same region rule as the main request? | Yes. The rule is enforced on every prompt-bearing call: the routing classifier, the answer, every retry and every fallback. Prompts are processed in memory and never written to storage or logs. | Security |
| DR-04 | Where is account, billing, usage and log metadata stored, and does any of it leave the region? | Your account and usage records are stored in Frankfurt, Germany, and our EU gateway runs in Finland. Request records hold metadata only (model, provider, tokens, cost and status), never your prompts or outputs. | SecuritySub-processors |
Retention and zero data retention
| ID | Question | Answer | Evidence |
|---|---|---|---|
| RT-01 | Does the gateway itself store prompts or completions, for how long, and can storage be switched off? | No. We process your prompts in memory to serve each request and never write them to storage or logs. | SecurityDPA section 3 |
| RT-02 | Can you require zero data retention (ZDR) from downstream model providers, and how is it enforced? | Yes. Turn on zero data retention for your company, a team or a key, and every request is routed only to inference endpoints that commit to keeping no copy of your data. Teams and keys inherit the requirement and can never switch it off underneath you. | SecurityGovernance |
| RT-03 | What exactly does the provider's ZDR commitment cover (abuse monitoring, caching, safety classifiers, human review)? | Every endpoint that serves zero-retention traffic must meet our Provider Requirements: prompts and outputs are not stored after the request is completed, and only transient caching strictly necessary for inference is accepted. | Provider Requirements |
| RT-04 | What happens to your data at contract end, and how quickly is it deleted? | Your content is never retained. On termination we delete or return all other customer personal data within 30 days. | DPA section 10 |
Training and secondary use
| ID | Question | Answer | Evidence |
|---|---|---|---|
| TR-01 | Does the vendor use your prompts, completions or files to train or improve models, and is that the default? | No. Standard Compute never trains on your prompts or outputs, and the DPA commits us to it. | DPA section 3.1 |
| TR-02 | Does the vendor require its sub-processors and downstream model providers not to train on your data, and does routing enforce that? | With a region rule or zero retention turned on, every request goes only to providers that commit not to train on it. | Provider RequirementsSecurity |
| TR-03 | Is any of your data used for analytics, benchmarking, product improvement or human review? | Prompts and outputs are processed in memory, never written to storage or logs, and never used for training. Request records hold metadata only: model, provider, tokens, cost and status. | Security |
Sub-processors
| ID | Question | Answer | Evidence |
|---|---|---|---|
| SP-01 | Is there a public, current list of sub-processors with purpose, data categories and location? | Yes. We publish every sub-processor with its purpose, data and location. | Sub-processors |
| SP-02 | How much advance notice do you get before a new sub-processor is added, and can you object? | We give you 14 days' notice before adding a new sub-processor. How to receive notices and object is set out on the sub-processor page. | Sub-processors |
| SP-03 | When the gateway adds a new model provider, does your traffic start reaching it automatically? | With a provider allow-list, a new provider is never added without an admin's approval, and fallbacks can never route around your list. | Sub-processorsGovernance |
Identity: SSO and SCIM
| ID | Question | Answer | Evidence |
|---|---|---|---|
| ID-01 | Which SSO protocols are supported (SAML, OIDC), with which identity providers, and on which plan? | Connect Microsoft Entra ID or any SAML identity provider so your people sign in to Standard Compute with their work accounts. Anyone on your verified domain joins your company automatically. | SSOSSO docs |
| ID-03 | Is SCIM provisioning supported, and does deprovisioning in the IdP revoke API keys and sessions? | Access to Claude Desktop and Claude Code follows your Microsoft Entra groups through SCIM provisioning. | Sign-in docs |
| ID-04 | Can access to AI tools (chat apps, coding agents) follow IdP groups rather than shared API keys? | Yes. Employees sign in to Claude Desktop and Claude Code with their Microsoft work account, through our EU gateway, so there are no API keys to hand out. For API use, every person and every app gets its own key. | SSOSign-in docs |
Audit logs and request records
| ID | Question | Answer | Evidence |
|---|---|---|---|
| AU-01 | Is there an admin audit log of configuration changes (people, keys, budgets, policies), with before and after values? | Yes. Every admin change is recorded with before and after values. | GovernanceSecurity |
| AU-02 | Is there a per-request record (model, serving provider, region, outcome, fallbacks) and can it be exported? | Yes. You can export a record of every request showing the model, the provider that served it, the region rule in force, the outcome and any fallback attempts reported by providers. | SecurityGovernance |
| AU-03 | How long are audit and request records kept, and can you choose a period? | Request records are kept for 12 months on Enterprise and the SLA add-on, and you can export them from the dashboard. | Security |
| AU-04 | Do audit records contain prompt content? If so, who can see them? | No. Request records hold metadata only (model, provider, tokens, cost and status), never your prompts or outputs. | Security |
Spend controls
| ID | Question | Answer | Evidence |
|---|---|---|---|
| SC-01 | Can you set hard budgets per key, per user, per team and per organization? | Every person and app gets a fixed monthly budget that never overruns, and you can set spending caps per key and for the whole company. | Budget docsGovernance |
| SC-02 | What happens when a budget is reached: hard stop, overage, or fallback to a cheaper model? | Hard stop, with no overage: requests return HTTP 402 with code budget_exhausted until the period renews. Optional top-off from a prepaid company balance keeps critical agents running. | Budget docs |
| SC-03 | Can spend be attributed to a person, app, team, project or cost code for chargeback? | Every person and every app has its own key and its own monthly budget, organized into teams under one company account. | Teams |
Model and provider governance
| ID | Question | Answer | Evidence |
|---|---|---|---|
| GV-01 | Can you restrict which models each team, key or app may use? | Yes. You decide which models each key can use, so every app and every team member gets exactly the models you approve. | Governance |
| GV-02 | Can you restrict which inference providers (clouds, hosts) may serve your traffic, and does the restriction also bind fallbacks? | Yes. Choose exactly which inference providers may handle your data, for the whole company, a team or a single key. Fallbacks can never route around your list. | GovernanceSecurity |
| GV-03 | Can lower levels (team, key) only narrow, never widen, the organization's policy? | Yes. Lower levels can only narrow the rules set above them. | Security |
Incidents, availability and assurance
| ID | Question | Answer | Evidence |
|---|---|---|---|
| IN-01 | What is the contractual breach notification window? | We notify you of a personal data breach within 72 hours of becoming aware of it. | DPA section 7Security |
| IN-02 | Is there a public status page and incident history, and an uptime SLA with credits? | Our status page is public, and Enterprise includes a 99.9% monthly uptime SLA with service credits. | StatusSLA |
| IN-03 | How does failover work when a model provider is down, and does it respect your region and provider rules? | If a provider goes down, your request fails over to another approved provider of the same model automatically, without leaving your region or provider rules. If no permitted route is available, the request fails and is never sent outside your policy. | Security |
EU AI Act deployer support
| ID | Question | Answer | Evidence |
|---|---|---|---|
| AI-01 | Can the service help you keep the logs a deployer of a high-risk AI system must keep (Art. 26(6): automatically generated logs under your control, kept for at least six months unless applicable Union or national law provides otherwise)? | Yes. You can export a record of every request with the model, serving provider, region rule and outcome, kept for 12 months on Enterprise and the SLA add-on. | SecurityGovernance |
| AI-02 | Does the vendor document which models it serves, who provides them, and links to provider documentation (model cards, usage policies)? | Every request record names the model and the provider that served it, and the models we serve are listed on our models page. | Models |
Exit and portability
| ID | Question | Answer | Evidence |
|---|---|---|---|
| EX-01 | Is the API compatible with a widely used interface (for example the OpenAI or Anthropic API format), so you can switch vendors by changing a base URL? | Yes. You send OpenAI- or Anthropic-compatible API requests to one endpoint, documented in our API reference. | API referenceSecurity |
| EX-02 | Can you export your configuration (keys, policies, prompts, routing rules) and your logs in a standard format? | Request records export from the dashboard, with the model, serving provider and region rule for every request. | Security |
| EX-04 | What are the termination terms: notice, data return, deletion, and any fees? | On termination we delete or return your personal data within 30 days. | DPA section 10Terms |
Contract and DPA
| ID | Question | Answer | Evidence |
|---|---|---|---|
| CT-01 | Is a Data Processing Agreement available, and can it be signed without a custom negotiation? | Yes. The full DPA text is public, and a company admin signs it in the dashboard in minutes. It commits us to no training on your data, no retention of your content, and breach notification within 72 hours. | DPA text |
| CT-02 | Which transfer mechanisms cover transfers outside the EU/EEA (SCCs, UK IDTA, EU-US Data Privacy Framework)? | The EU Standard Contractual Clauses and, for UK data, the UK International Data Transfer Addendum. The mechanism for each sub-processor is listed on our sub-processor page. | DPA section 6Sub-processors |
| CT-03 | Which law governs the contract, and where are disputes resolved? | Norwegian law. Disputes go to the courts of Oslo under our terms, and Oslo District Court is the agreed venue under the DPA. The contracting entity is Standard Compute AS, org. no. 938 498 806, Oslo, Norway. | TermsDPA section 11 |
About the questionnaire
The questionnaire is open and vendor-neutral, MIT-licensed, and written for evaluating any LLM gateway or AI API provider. The question IDs above match its spreadsheet, so each answer pastes into the matching row. For the full procurement view, with data categories, locations and DPA terms in one place, see our security review pack.
Create your company workspace to sign the DPA, connect your identity provider and set your region and provider rules. For a questionnaire in your own format, write to us and we return it completed.