AI Tools

Calculators for building with large language models (LLMs): tokens, context windows, RAG, costs, ROI and hardware. They run in your browser and use the numbers you enter (prices, token counts), so they stay correct as models and prices change. Full list · Detailed guide for every tool.

Always enter current prices from your provider's pricing page. Token estimates (from words or characters) are approximations; use your provider's tokenizer for exact counts.

Groups#

GroupToolsWhat for
Tokens & context8Words ⇄ tokens, characters → tokens, file token estimate, Context Window Calculator (what is left after system prompt, history and reply), context usage (how many more turns fit), max output (words/pages a max-token setting allows)
RAG & embeddings6Chunk count (with overlap), chunk size (largest chunk such that top-k results fit the prompt), knowledge-base size (tokens, chunks, vector storage), embedding cost, vector storage / memory, RAG ROI
Cost & pricing14LLM API Cost (per request/day/month), prompt-caching savings, batch API savings, model-vs-model cost comparison, fine-tuning cost, infrastructure cost, self-host vs API, usage-based pricing, AI-SaaS pricing, margin, MRR, LTV, CAC, unit economics
ROI & savings11AI ROI, automation savings, customer-support savings, chatbot ROI, agent ROI, productivity gain, cost reduction, workforce savings, sales-agent ROI, marketing ROI, revenue impact
Hardware & speed4VRAM Calculator (GPU memory for a model at a quantisation), tokens per second, GPU rental cost, training time

Worked examples#

1. "Will my prompt fit?" (Context Window Calculator)#

Window 200,000 tokens, system prompt 3,000, chat history 40,000, reserved reply 8,000 → 149,000 tokens left for documents.

2. "How much will this cost?" (API Cost)#

10,000 requests/day × (1,500 input + 400 output tokens) at your provider's per-million-token prices → cost per request, per day, per month. Change one number to see the effect of a shorter prompt.

3. "Does caching pay off?"#

Prompt with a long fixed prefix (a policy document, a codebase summary) → Prompt Caching Savings shows the monthly saving when the prefix is cached on repeated calls.

4. "API or my own GPU?" (Self-host vs API)#

Enter monthly token volume, API prices, GPU rental/hardware cost, utilisation → break-even volume.

5. "What GPU do I need?" (VRAM Calculator)#

A 70 B-parameter model at 4-bit quantisation needs roughly 35 GB of weights plus overhead for the context cache; the tool adds the KV-cache for your context length and tells you what fits on 24 GB / 48 GB / 80 GB.

6. RAG sizing#

Document set of 5 million tokens, chunk 500 tokens with 50 overlap → chunk count, embedding cost, and vector storage (dimension × bytes × chunks), to choose between a managed vector DB and a local index.

Tips#

  • Estimate input and output separately: output tokens usually cost several times more.
  • Remember hidden tokens: system prompts, tool definitions, retrieved context and conversation history all count.
  • Batch (async) and caching discounts can cut cost a lot, but depend on provider rules.
  • Business calculators (ROI, LTV, CAC, MRR) are only as good as your assumptions: show best/worst cases.