Local vs cloud RAG cost calculator

Plug in your documents and how much you ask, then compare hosted file search from OpenAI, Gemini, Claude and Grok with running it on your own hardware. Cloud prices come from official pricing pages (checked 2026-10-27). Hardware prices are editable estimates; put in what you'd really pay.

1. Your documents and usage

Number of documents Avg pages per document Tokens per page (≈500 for text) Questions per month Question tokens Retrieved context tokens per question Answer tokens Full re-indexes per month Retrieval calls per question

2. Cloud models

3. Local hardware

Hardware Upfront cost ($, editable estimate) Average power draw (W) Busy hours per day Idle power (W) Electricity $/kWh Amortize over (months)

Monthly cost

Option	Monthly	Breakdown	First index (one-time)	Break-even vs local

Break-even = hardware upfront ÷ (cloud monthly − local electricity). "Never" means the cloud option costs less than your power bill. Storage size assumes about 4 bytes of text per token (an estimate). Excludes taxes, free tiers (except OpenAI's 1 GB), batch discounts and caching.

Next steps

Going local? Pick a tool: all tools, by category, by hardware.
Which model fits your GPU: local models. On a Spark: DGX Spark guide.
Policies and limits side by side: local vs cloud. Want Claude/Cursor to search your local docs: MCP tools.