Chat Prompt Token Estimator

Overview

You send the messages, tool schemas and model profile you are about to use, and we tell you how many tokens the whole request comes to, how much of your context window is left, and what it would cost at a price you supply. The content half is exact: we run the real tokenizer over your message text, the same one behind text.count_tokens, so the two can never disagree. The framing half is an estimate, and we say so in the response with a literal estimate flag — a chat request costs more than the sum of its message bodies, because each message carries role and delimiter framing and the reply is primed, and providers can change that wire format without notice. We ship no price list and no context-window list: both are yours to supply, because a table baked into a skill is wrong the week a provider changes it and a confidently wrong cost is worse than no cost. If all you want is the token count of one string, text.count_tokens is the smaller, cheaper answer.

Token surprises are expensive and late. A conversation that fits in testing overflows in production once a tool schema and a few turns of history join it, and the first sign is a truncated answer or a rejected call. Getting the number locally, with the envelope included and the arithmetic published, turns that into a decision you make before you pay for it.

When to use it

Scope

Non-goals

• Does not call a model, a tokenizer service or any inference API
• Does not ship a price list — every rate comes from you
• Does not ship context-window sizes — that number is an input, not a lookup
• Does not count a plain string: text.count_tokens is the skill for that
• Does not bill anything; the estimate flag in the answer says so out loud

Properties

Capabilities

Cost & billing

Flat rate per invocation

File handling

How we handle your data

Input schema

Output schema

API contract

curl -sS -X POST "$BASE/v1/execute" -H "Authorization: Bearer $KEY" -H 'Content-Type: application/json' -d '{"skillKey":"ai.prompt_token_estimate","input":{"messages":[{"role":"system","content":"You are a support agent."},{"role":"user","content":"Where is order NW-84120?"}],"modelProfile":{"contextWindow":8192}}}'

import { Squillify } from "@squillify/sdk";

const sq = new Squillify({ apiKey: process.env.SQUILLIFY_API_KEY });

const { output } = await sq.execute("ai.prompt_token_estimate", {"messages":[{"role":"system","content":"You are a support agent for Northwind Supplies. Answer in British English and never promise a delivery date."},{"role":"user","content":"Order NW-84120 has not arrived. What should I tell the customer?"}],"modelProfile":{"contextWindow":8192,"reservedOutputTokens":1000},"pricing":{"inputPricePerMillion":"2.50","outputPricePerMillion":"10.00","currency":"EUR","expectedOutputTokens":400}});

Errors

Other limits

Limitations

• Does not call a model, a tokenizer service or any inference API
• Does not ship a price list — every rate comes from you
• Does not ship context-window sizes — that number is an input, not a lookup
• Does not count a plain string: text.count_tokens is the skill for that
• Does not bill anything; the estimate flag in the answer says so out loud
• Counting the tokens of one plain string, which text.count_tokens does directly
• Producing a billable figure — this is a planning number, never an invoice
• Models whose tokenizer is not cl100k_base or o200k_base

How to call

POST /v1/execute (requires authentication)

{
  "skillKey": "ai.prompt_token_estimate",
  "input": {
    "messages": [
      {
        "role": "system",
        "content": "You are a support agent for Northwind Supplies. Answer in British English and never promise a delivery date."
      },
      {
        "role": "user",
        "content": "Order NW-84120 has not arrived. What should I tell the customer?"
      }
    ],
    "modelProfile": {
      "contextWindow": 8192,
      "reservedOutputTokens": 1000
    },
    "pricing": {
      "inputPricePerMillion": "2.50",
      "outputPricePerMillion": "10.00",
      "currency": "EUR",
      "expectedOutputTokens": 400
    }
  }
}

Examples

{
  "messages": [
    {
      "role": "system",
      "content": "You are a support agent for Northwind Supplies. Answer in British English and never promise a delivery date."
    },
    {
      "role": "user",
      "content": "Order NW-84120 has not arrived. What should I tell the customer?"
    }
  ],
  "modelProfile": {
    "contextWindow": 8192,
    "reservedOutputTokens": 1000
  },
  "pricing": {
    "inputPricePerMillion": "2.50",
    "outputPricePerMillion": "10.00",
    "currency": "EUR",
    "expectedOutputTokens": 400
  }
}

{
  "estimate": true,
  "tokens": {
    "content": 37,
    "framing": 11,
    "tools": 0,
    "total": 48
  },
  "perMessage": [
    {
      "index": 0,
      "role": "system",
      "contentTokens": 21,
      "framingTokens": 4
    },
    {
      "index": 1,
      "role": "user",
      "contentTokens": 16,
      "framingTokens": 4
    }
  ],
  "headroom": {
    "contextWindow": 8192,
    "reservedOutputTokens": 1000,
    "availableTokens": 7144,
    "fits": true,
    "overflowTokens": 0
  },
  "cost": {
    "inputCost": 0.00012,
    "outputCost": 0.004,
    "totalCost": 0.00412,
    "currency": "EUR",
    "rounded": 0
  },
  "framing": "CHATML_V1",
  "tokenizer_version": "jtokkit-1.1.0",
  "tokenizer": "cl100k_base"
}

{
  "messages": [
    {
      "role": "user",
      "content": "Find the shipping status for order NW-84120."
    }
  ],
  "pricing": {
    "currency": "JPY",
    "inputPricePerMillion": "150",
    "expectedOutputTokens": 0
  },
  "modelProfile": {
    "tokenizer": "o200k_base",
    "framing": "CHATML_TOOLS_V1",
    "contextWindow": 128000
  },
  "tools": [
    {
      "name": "lookup_order",
      "parametersJson": "{\"type\":\"object\",\"properties\":{\"orderId\":{\"type\":\"string\"}},\"required\":[\"orderId\"]}",
      "description": "Look up an order by its reference and return its shipping status."
    }
  ]
}

{
  "estimate": true,
  "tokens": {
    "content": 11,
    "framing": 7,
    "tools": 43,
    "total": 61
  },
  "perMessage": [
    {
      "index": 0,
      "role": "user",
      "contentTokens": 11,
      "framingTokens": 4
    }
  ],
  "headroom": {
    "contextWindow": 128000,
    "reservedOutputTokens": 0,
    "availableTokens": 127939,
    "fits": true,
    "overflowTokens": 0
  },
  "cost": {
    "inputCost": 0.00915,
    "outputCost": 0,
    "totalCost": 0.00915,
    "currency": "JPY",
    "rounded": 0
  },
  "framing": "CHATML_TOOLS_V1",
  "tokenizer_version": "jtokkit-1.1.0",
  "tokenizer": "o200k_base"
}

Recommended use cases

• Decide whether a conversation still fits before you send it, and how much room is left
• Compare what a tool schema adds to every turn against what it buys you
• Put a euro figure beside a prompt design review, at your own negotiated rate

Avoid for

• Counting the tokens of one plain string, which text.count_tokens does directly
• Producing a billable figure — this is a planning number, never an invoice
• Models whose tokenizer is not cl100k_base or o200k_base

Related skills

Composable with

Safety & privacy

• Runs without calling any external service.
• Your payloads are never stored permanently.
• Telemetry captures metadata only — never your payload contents.
• We never train models on your payloads.
• No human reviews your payloads.

Tags