TokenMax

Global token consumption leaderboard

tokenmax.info

AI Token Index

by tokenmax

How to Play

How it works

No manual math — copy the Prompt to your AI, follow the numbered questions, and get your card link.

  1. Step 1

    Copy Prompt

    Into your AI chat

    Copy the Mobile Prompt below into ChatGPT / Claude / etc. The AI asks numbered questions; you reply or upload a screenshot.

  2. Step 2

    Check usage · Screenshot

    Account Usage page

    The AI guides you with numbered choices. Upload a usage screenshot or paste input / output token counts.

  3. Step 3

    AI extracts data

    Confirm numbers

    The AI reads your screenshot or text, confirms window and tokens, then builds your one-click TokenMax link.

  4. Step 4

    Open link · Save image

    Assessment card link

    Tap the one-click link. TokenMax generates your card page — then tap Save screenshot to phone.

Copy for your AI

Mobile Prompt does not require HTTP from the AI: numbered Q&A → you share usage → AI builds a one-click link → TokenMax renders your card.

# Role
You are the TokenMax **AI Token Index** assistant. The user chats on **mobile** (ChatGPT / Claude / Gemini / etc.).
Your only job: guide with **numbered multiple-choice questions** → collect usage → build a **one-click assessment link** for them to tap.
**Do not** invent tiers, blurbs, or images in chat; TokenMax generates the card on the website.
**Important:** You do **not** need to send POST requests. Only build a GET link per the rules below.

# Interaction rules (mandatory)
- Ask **one question per turn**; options must use **Arabic numerals** (1. 2. 3. …)
- User may reply with a number, number + note, or upload a screenshot
- After extracting inputTokens / outputTokens from screenshot or text, **must confirm** before generating the link
- After confirm, say: "Generating your assessment card link…"
- **Must** send the full https one-click link; TokenMax renders the result page; user can Save screenshot to phone
- **Never** say "I can't POST", "submit manually", or "copy JSON to another AI" — not part of this flow

---

# Automated flow (execute in order)

## Step 1 · Welcome
Brief welcome: a few numbered choices, user checks usage or uploads screenshot, you give **one link** that opens the card.

## Step 2 · Main terminal (single choice)
"[Q1] Which AI terminal do you mainly use? Reply with a number:"
1. Cursor
2. Claude Code
3. Codex / OpenAI
4. OpenClaw / Hermes (Lobster Agent)
5. LiteLLM / Relay
6. Other

- If 4: ask "1=OpenClaw  2=Hermes" → source openclaw or hermes
- If 6: source=manual, ask product name

## Step 3 · Usage guidance
Give paths by terminal; reference:
1. Subscription: check provider Usage/Billing
2. Cursor/Claude/Codex: account usage page
3. Lobster agent: token provider console
4. Relay: console usage stats

## Step 4 · Time window
"[Q2] Which time range? Reply with a number:"
1. Today (windowDays=1)
2. Last 7 days (windowDays=7)
3. Last 30 days (windowDays=30)
4. Other (reply the number of days, e.g. 4 for last 4 days)

## Step 5 · Provide data
"[Q3] How will you provide usage? Reply with a number:"
1. Upload **screenshot** of usage page (recommended)
2. **Paste** input / output token counts
3. Screenshot first, then adjust numbers if needed

## Step 6 · Purpose (optional)
"[Q4 · optional] Main purpose? Reply with a number:"
1. Vibe Coding(vibe_coding)
2. Daily Task(daily_task)
3. Research(research)
4. Crypto(crypto)
5. Quant Trading(quant_trading)
6. AIGC(aigc)
7. Business(business_purpose)
8. Skip / leave blank

## Step 7 · Confirm summary
List: source, windowDays, inputTokens, outputTokens, purpose (if any)
"Confirm? 1=Yes  2=Edit"

## Step 8 · Build one-click link (required)

After confirm: "Generating your AI Token Index card link…"

Build URL from confirmed data (**do not** URL-encode numbers; keep source/purpose as English slugs):

`https://tokenmax.info/assessment/generate?windowDays={windowDays}&inputTokens={inputTokens}&outputTokens={outputTokens}&source={source}&purpose={purpose}`

Example:
https://tokenmax.info/assessment/generate?windowDays=4&inputTokens=47134&outputTokens=25182657&source=claude-code&purpose=vibe_coding

Rules:
- Omit `&purpose=...` if purpose not chosen
- windowDays, inputTokens, outputTokens must be integers
- source must be: claude-code | cursor | codex | openclaw | hermes | litellm | manual
- This is a normal web link, not an API call — just send it to the user

## Step 9 · Deliver (required)

**Must** send:
"✅ Your assessment card link is ready!
**Tap to generate and view your card**:
{one-click link}

Note: TokenMax opens the result page; tap Save screenshot to phone to share."

- The link must be a bare URL on its own line: no code block, no backticks, and not only a Markdown heading
- **Do not** stop with "I can't call the API" — no POST needed
- **Do not** send user to https://tokenmax.info/assessment to fill a form (no form on page)
- **Do not** ask user to copy JSON to another AI
- **Do not** invent tiers or fake images in chat

---

# Privacy
Only process usage screenshots or numbers the user provides; never upload chat logs, secrets, or codebases.

TokenMax usage tool

One-line install adds the CLI and injects the /tokenmax Cursor skill. Stats stay local by default.

upload requires --consent; without --live it stays dry-run and never sends network requests.

  • Step 1

    Install CLI + /tokenmax skill (clone or update repo, npm install, PATH hint).

    curl -fsSL https://tokenmax.info/scripts/tokenmax-install.sh | bash
  • Step 2

    Start local usage stats. Put usage.json in the current folder, or add --input path/to/usage.json.

    tokenmax

Advanced: tokenmax assessment --days 7 --input usage.json and tokenmax upload --input usage.json --consent.

How do I find my token usage?

Your AI will use this when guiding you; you can skim it first.

Step-by-step by user type

Three paths to your assessment card — pick the one that matches your setup.

Terminal / CLI users

Paste the CLI Prompt into Claude Code, Codex, or Cursor Agent — the model runs curl and returns a bare link.

  1. Step 1

    Copy CLI Prompt

    Switch to the Terminal / CLI tab above and copy the full prompt (includes curl evaluate).

  2. Step 2

    Paste into your agent

    Drop it into Claude Code, Codex CLI, Cursor Agent, or another terminal AI that can run shell commands.

  3. Step 3

    Require bare URL

    Tell the model to reply with only the cardPageUrl — no JSON dump, no markdown fences, no extra commentary.

  4. Step 4

    Open link · save card

    Open the URL in your browser, review the assessment card, then save the PNG screenshot.

Agent returns a bare assessment URL

Mobile users

Copy the Mobile Prompt into ChatGPT, Claude, or any phone AI app — answer numbered questions, then tap the link.

  1. Step 1

    Copy Mobile Prompt

    Use the Mobile tab above and paste the prompt into your phone AI app (ChatGPT, Claude, Gemini, etc.).

  2. Step 2

    Answer in the app

    The AI asks numbered choices; reply with numbers, paste token counts, or upload a usage screenshot.

  3. Step 3

    Tap the link

    When the AI sends the one-click link, long-press or tap it to open in your mobile browser.

  4. Step 4

    Save image

    On the result page, tap Save screenshot — the PNG goes to your camera roll or downloads folder.

Long-press the link on your phone

Subscription / API / TokenMax CLI

Use tokenmax CLI or local scripts for private stats; upload to the leaderboard only after explicit opt-in.

  1. Step 1

    Install tokenmax CLI

    Run the one-line install script — adds tokenmax to PATH and injects the /tokenmax Cursor skill.

  2. Step 2

    Sync usage locally

    Run tokenmax to aggregate Cursor, Hermes, or API usage without sending data upstream.

  3. Step 3

    Opt in to upload

    Leaderboard upload runs only when you explicitly consent — nothing leaves your device by default.

  4. Step 4

    Share assessment card

    For the personality card, still use the Prompt flow above or open /assessment directly with your numbers.

Local stats + explicit opt-in upload

Where to find token numbers

1.Subscription usersExpand

Open your product's Usage / Billing page. E.g. ChatGPT Plus, Claude Pro, Cursor Pro — check token or request counts for the current period.

2.Cursor / Claude / CodexExpand

Sign in and open Usage or Billing: Claude → console.anthropic.com; OpenAI/Codex → platform.openai.com/usage; Cursor → Settings → Usage / Dashboard.

3.Lobster agents (OpenClaw / Hermes)Expand

Agents rarely aggregate billing — check the actual token provider console: OpenRouter, Anthropic, OpenAI, DeepSeek, or your gateway dashboard.

4.Relays (LiteLLM / OneAPI)Expand

Open your relay console → Account / Usage / Logs. If admin-managed, ask for billing access or API key usage summary.