Enterprise AI glossary
Short definitions of the terms that come up when a company buys, secures and governs AI: gateways and routing, data protection, identity, spend control and regulation.
Each entry gives the definition first, then why it matters, a concrete example and how Standard Compute handles it.
Gateways, routing and capacity
- AI gateway
- An AI gateway is a single service that sits between a company's applications and the AI model providers it uses.
- LLM router
- An LLM router is software that decides, for each request, which large language model and which provider should answer it.
- Model routing
- Model routing is the practice of sending each AI request to a model chosen for that request, instead of sending everything to one fixed model.
- Tokens per second
- Tokens per second (TPS) is the speed at which a model generates its answer, measured as output tokens per second for a single request.
- Rate limit vs quota
- A rate limit caps how fast you can call an AI model, for example requests or tokens per minute.
- Provisioned throughput (PTU)
- Provisioned throughput is reserved model capacity: you pay for a fixed amount of processing per hour, whether or not you use it, in return for predictable throughput and latency.
Data protection and location
- Zero data retention (ZDR)
- Zero data retention (ZDR) means the AI provider keeps no copy of prompts or outputs after the response is returned.
- No-training guarantee
- A no-training guarantee is a contractual commitment that an AI provider will not use your prompts, files or outputs to train or improve its models.
- Data residency
- Data residency is the requirement that data is stored, and often processed, in a specific country or region.
- Regional processing (Europe rule)
- Regional processing is a routing rule that lets AI requests be served only by inference endpoints in an approved region.
- BYOK (bring your own key)
- BYOK (bring your own key) means you supply your own keys to a third-party tool.
- Sovereign AI
- Sovereign AI means a company or public body can use AI under its own jurisdiction and control.
Identity and access
- SSO, SAML and Microsoft Entra for AI tools
- Single sign-on (SSO) for AI tools lets employees sign in to AI services with their company identity, such as a Microsoft Entra ID work account, instead of a separate password.
- SCIM
- SCIM (System for Cross-domain Identity Management) is an open standard for creating, updating and removing user accounts and groups in an application automatically, driven by the identity provider.
- Provider allow-list
- A provider allow-list is the list of inference providers, such as a model maker's own API or a specific cloud, that may process a company's AI requests.
- Model allow-list
- A model allow-list defines which AI models a person, team or application may call.
Spend control
- Per-key budget
- A per-key budget is a spending limit attached to a single API key, so the person or application using that key can never spend more than its allowance in a period.
- Spend cap
- A spend cap is a hard upper limit on AI spending for an account, a key or a period.
Governance and regulation
- Audit log for AI
- An audit log for AI is a record of who changed AI settings and how AI was used: admin changes to people, keys and rules, and for each request, which model and which provider served it under which policy.
- AI usage policy
- An AI usage policy is a company's written rules for using AI tools at work: which tools are approved, what data may go into prompts, how outputs are reviewed and whom to tell when something goes wrong.
- EU AI Act deployer
- Under the EU AI Act, a deployer is a natural or legal person, public authority, agency or other body using an AI system under its authority, except in the course of a personal non-professional activity (Article 3(4)).
Standard Compute is an AI gateway for companies, with frontier and open-source models through one API. Your admins decide where requests are processed, which providers and models are allowed and what each person and app can spend.
Create a company workspaceor write to contact@standardcompute.com