Help · Glossary
Plain-English definitions for every metric, status, and concept in GreyScape.ai. Each entry has a literal definition plus a one-paragraph “what it means for your business.”
Month-to-date spend — total billed AI cost since the 1st of this month.
The headline number on the dashboard. Used to compare against your monthly budget and to forecast month-end total. A 30%+ week-over-week change in MTD spend is what triggers our default anomaly threshold.
See also: Budget utilisation, Forecast EOM
End-of-month forecast — projected total spend by the last day of the current month.
MTD spend × (days in month ÷ day of month). A naive but transparent forecast that updates daily.
See also: MTD spend, Budget utilisation
Forecast EOM as a percentage of the active monthly budget.
57% on track. 75% triggers a soft alert. 100% triggers a hard alert and (in Phase 3) blocks new approvals.
Total cost for a workload divided by number of invocations.
Lets you compare the unit economics of two workloads. A classification job at $0.003/call is often acceptable; a chat completion at $0.30/call demands more scrutiny.
A workload that's been scoped by a requester and is awaiting (or has received) a decision from an admin.
The core unit of governance. Every new AI use case in your org should flow through one. Approval auto-creates a budget and (for OpenAI/Anthropic) auto-provisions an access key tagged to the requester.
See also: Approval status, Budget scope
draft → scoping → submitted → approved | rejected.
Draft: link created but no chat yet. Scoping: requester is chatting with the advisor. Submitted: ready for admin decision. Approved: monthly budget active, provisioning available. Rejected: closed, requester sees the decision notes.
The level at which a budget caps spend: organisation, team, user, or project.
Organisation: total monthly AI spend across the company. Team: per-team cap (cost centre). User: per-employee cap. Project: per-workload cap (created automatically on each approval).
On approval, GreyScape.ai calls the provider's admin API to create a project and scoped key.
Removes the manual step where an admin has to log into OpenAI / Anthropic to generate a key for the requester. Currently supported: OpenAI, Anthropic (workspace; key still manual). Azure, Bedrock, Google in roadmap.
Live monthly spend on an approved workload compared to its approved budget.
The closed-loop metric. Each approval auto-creates a project-scoped budget; spend events tagged with the same project name update the actual figure on the request page within 15 minutes of every sync. Tells you whether each approval is on track.
Spend pattern that deviates from the rolling 14-day baseline by more than 3 standard deviations.
Statistical flag, not always a problem — could be a legitimate new workload. Severity is graded: danger (likely runaway agent), warn (significant pattern shift), info (worth knowing, not urgent).
See also: Anomaly severity
danger | warn | info.
Danger: a workload's daily spend exceeded its 14-day baseline by >5σ — review within 24 hours. Warn: 3–5σ deviation — review this week. Info: between 2 and 3σ — worth noting but no urgent action.
AI tools used by employees outside the sanctioned organisation account.
Personal ChatGPT Plus on corporate expense cards. Cursor signups via Google SSO. Local LLM tools on laptops. Each represents an unbudgeted spend AND a data-leakage risk. The shadow-AI dashboard surfaces them via expense, SSO, MDM, and network feeds.
review | sanctioned | blocked | restricted.
Review: detected but no admin decision yet. Sanctioned: admin has formally approved it. Blocked: admin has marked it disallowed. Restricted: allowed for a defined subset of users only.
Organisation-level credential that reads usage and cost across every project in the provider org.
Different from a regular API key. OpenAI's are prefixed sk-admin-; Anthropic's sk-ant-admin01-. You generate one per provider, not one per user. It's the only credential GreyScape.ai needs to attribute every employee's usage.
Bring-your-own-key — your own provider API keys, tracked by GreyScape.ai.
GreyScape.ai never resells or proxies AI traffic. You keep your direct relationship with OpenAI / Anthropic / others; we just observe your usage via their admin APIs. No data passes through our servers in the request path.
A scheduled call to a provider's admin API that pulls the last 14 days of usage and cost.
Runs every 15 minutes via Railway cron. Each run is logged with success / error status. If your dashboard looks stale, check the Last sync time on /connectors/providers.
live | stale | erroring | pending.
Live: synced successfully in the last 30 minutes. Stale: no successful sync for >30 minutes (data may be out of date). Erroring: last attempt failed. Pending: credential stored, no sync run yet.
An auto-provisioned, project-scoped key created when a request is approved.
Lives inside a specific OpenAI project named after the approval. The project has a hard monthly spend cap matching the approved budget. The key is shown once on the request detail page; archive it in a password manager and share with the requester.
Allowlist of model IDs the AI advisor is permitted to recommend.
Use this when your org has data-residency or vendor-contract constraints. The advisor sees only the models in the list — it cannot suggest a disallowed model even if it's the cheapest fit. Empty list = advisor refuses to recommend anything.
Missing a term? Email [email protected] and we’ll add it. · Public glossary