AI Savings & Capacity Calculator
Two modes, matching the two ways VDF AI is licensed. Cloud is a managed subscription priced per user. On-premises is Capacity Licensing — an annual subscription sized to the work the platform performs, with unlimited people. Below: how each mode calculates, what the assumptions are, and how to embed the tool.
Live calculator
Same build as the standalone file — switch tabs to move between the two licensing models.
Cloud subscription mode
VDF AI Cloud is a managed subscription priced per user. Six inputs: team size, and five monthly spend lines (LLM API, agent/orchestration, ML/evaluation, chat/productivity seats, other tooling). The five lines are summed and annualized (×12) into current annual spend.
Why there is no list price in the tool
VDF AI publishes the structure of its pricing, not a rate card. Per-user rates are confirmed on a quote, so the calculator does not invent one. Instead it reports the break-even rate — your annualized stack spend divided by users and by twelve. That is the number a CFO actually needs: any per-user rate below it consolidates the stack at lower cost than the status quo. Enter a quoted rate and the tool models the annual saving against it directly.
Three plans are shown for context — Starter (free to start), Professional (per user), and Enterprise Cloud (annual, scoped to your rollout). The comparison is deliberately about consolidation: a multi-vendor stack of cloud APIs, orchestration tooling, an ML platform, and copilot seats, versus one governed platform.
Model-provider token spend and the infrastructure running the models stay separate in every scenario. Routing efficiency is modeled in the on-premises tab, not here, so the two effects are never double-counted.
On-premises Capacity Licensing mode
On-premises VDF AI is licensed by capacity, not by headcount. Named users, API keys, service accounts, agents and workflows are unlimited in every bundle. What you license is the volume of work the platform performs during the term. The mode runs in three steps.
Step 1 — the token meter
Interactive volume is people × interactions per day × tokens per interaction ×
working days, with presets for occasional, regular, and intensive profiles. Retrieval
indexing is added separately as documents × tokens per document, because
embedding tokens generated at indexing time are counted once.
Step 2 — the transaction meter
A platform transaction is one inbound request to VDF AI, counted once however many rules, integrations, steps, or model calls run inside it. The tool sums API calls, scheduled job executions, and triggered workflow runs per day, adds one transaction per 100 batch records or part thereof, and annualizes over the days automation actually runs.
Step 3 — which meter binds
Both meters run in parallel, and a bundle reaches its limit when either allowance is reached. Because capacity bundles are balanced at roughly a thousand tokens per transaction, the binding meter is whichever side of your workload departs from that balance. Conversational deployments — many tokens per request, few requests — reach the token allowance first. Automated deployments such as fraud scoring, integration hubs, and straight-through processing reach the transaction allowance long before the token one, which is exactly the case where licensing by user count produces a meaningless number.
What never counts
Interactive sessions by your people; governance and policy evaluation, audit logging, and compliance reporting; administration, configuration, monitoring, and health checks; retrieval from an existing index inside a counted transaction; failed requests and automatic retries within 60 seconds of a failure. A streamed response counts once, at the request that initiated it.
The economics panel
Two independent effects. Unlimited users retires per-seat AI tool spend outright — and the growth slider shows what those seats would have cost by the end of the term as adoption grows, which under a Capacity Bundle is still nothing. Separately, routing and consolidation reduce what the infrastructure underneath costs to run: energy (your 30–65% routing band), DevOps and ML engineering 45%, model evaluation 80%, orchestration licences 60%, downtime and incidents 50%, and no modeled reduction on the "other" line.
Each row ties a mechanism to a percentage so finance can audit the logic rather than accept a headline. The Capacity Bundle itself is a separate fixed annual line, agreed before the term starts rather than invoiced after it — annual list pricing comes from VDF AI or your authorized partner. The output here is an estimate to bring to a sizing session, not a quote.
Embed the calculator
The tool is a single self-contained file with no external dependencies. Drop this iframe into any page:
The frame posts its own height to the parent window as it recalculates, so a host page can
size it exactly by listening for a vdf-calc-height message rather than
guessing a fixed height.