> ## Documentation Index > Fetch the complete documentation index at: https://docs.promptshields.com/llms.txt > Use this file to discover all available pages before exploring further. # The five views > Overview, Agents, Alerts & Budgets, People, and Connectors — and the question each one answers. Five tabs, each answering one question a finance or platform lead actually asks. ## Overview — *"What are we spending, and is anything wrong?"* The landing view. An alert ticker across the top, a 90-day stacked trend beneath it, then provider bars and team budgets. The trend is stacked by **spend layer**, so cloud, seats, and bespoke API usage are visually separable rather than collapsed into one number. A spike in one layer stays legible instead of being averaged away. Lower down, provider bars show concentration — which vendor owns what share of the bill — and team budget bars show consumption against each team's monthly allowance. ## Agents — *"Which agent is misbehaving?"* One card per agent, with its spend, its normal daily rate, and its current multiple of that rate. This is where a runaway loop becomes obvious: an agent running at several times its median daily cost is badged as an anomaly. The multiplier that triggers it is configurable rather than fixed, because "abnormal" differs by workload. ## Alerts & Budgets — *"What should I do about it?"* Two halves. The upper half is where you tune the rules: each team's monthly budget, the budget-warning threshold, and the agent-anomaly multiplier. Alerts are **computed, not seeded**. Change a budget or a multiplier and the ticker, the KPI counts, the budget bars, and the anomaly badges all recompute immediately. Settings persist in your browser, with a reset button. The lower half carries **savings recommendations** — also computed rather than hardcoded. Each is derived from actual usage multiplied by the list-price delta between the model in use and a cheaper tier. A recommendation to route a summarisation workload to a mini-tier model comes with the arithmetic behind it, so it can be argued with. ## People — *"Who is this spend attributable to?"* Every seat and every API key rolls up to a person, filterable by team. Click through for that individual's breakdown. This is the view that ties the three layers together: the same person may hold a Copilot seat, own an API key funding an agent, and sit in a team with a cloud budget. Per-person attribution **across all three layers at once** is the part that other tooling generally does not attempt. ## Connectors — *"Where does the data come from, and can I trust it?"* Sources grouped into three tiers by how obtainable the data is. The tiering is deliberately honest about a real constraint: several vendors keep per-member data locked in an admin console with no API. Rather than pretending otherwise, the Connectors view names the wall and shows the import path around it. What the Anthropic connector actually pulls, how the ChatGPT Enterprise CSV import works, and what each is allowed to claim. ## Theme Light and dark, both drawn from a CVD-validated categorical palette restepped for dark surfaces. The three spend layers map to fixed palette slots, so a layer keeps its colour across every chart and both themes — the colour is part of the data encoding, not decoration. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.