About AI Budget

Why it exists

Costing an LLM application is mostly arithmetic, but the arithmetic is spread across provider pricing pages, tokenizer behaviour, chunking decisions, and infrastructure line items. It is easy to get an answer that is wrong by an order of magnitude — usually by forgetting that retrieved context is billed as input tokens on every single request.

AI Budget puts those pieces in one place. The budget planner walks through application type, usage, model, RAG, and infrastructure, then shows a monthly estimate with the breakdown visible. The individual calculators exist for when you only need one number.

How the estimates are produced

Every calculation is a pure function running in your browser. Nothing is hidden behind an API, and no result is generated by a language model. The formulas are published on each tool page so you can check them by hand, and the planner lists the assumptions behind any figure it shows.

The core cost model is deliberately simple:

• LLM cost — monthly requests multiplied by input and output tokens per request, priced against the selected model's per-million-token rates.
• Embedding cost — document tokens turned into chunks, embedded once on ingest, plus any re-indexing you configure.
• Infrastructure — vector database, storage, and hosting, entered as ranges rather than a single invented number.

Token counts are estimates. Exact tokenization depends on the tokenizer the model actually uses, which is why the token calculator labels its output as an estimate rather than an exact count.

Where pricing comes from

Model pricing lives in a single configuration file rather than being scattered through the interface. Each entry records the provider, the rate, the date it was checked, and a link to the provider's own pricing page.

Pricing was verified against each provider's official pricing page on the date shown per model, then converted to INR at an approximate market exchange rate. Provider pricing changes without notice — confirm current rates before making a purchasing decision.

If you spot a rate that has moved, tell us — a correction is a one-line change to that configuration file.

What it does not do

• It does not guarantee costs. Real spend depends on traffic patterns, caching, retries, and provider changes. Treat every figure as an estimate to sanity-check, not a quote.
• It does not price every provider or every infrastructure vendor. The model list covers common choices; you can override pricing manually when your rate differs.
• It does not account for negotiated enterprise rates, committed-use discounts, free tiers, or taxes.
• It does not require an account, and there is no paid tier in this version.

Privacy

Because the calculations are client-side, the text and numbers you enter stay in your browser. Nothing you type into a calculator is sent to a server or stored by us. See the privacy page for the full detail.

Learning the concepts

If a term in the planner is unfamiliar, the guides cover the underlying ideas — tokens, how LLM pricing works, RAG, and context windows — with worked examples rather than definitions alone.