⚡ THE TOKENDOME

BURN TOKENS
CLAIM GLORY.

Live leaderboard for LLM token burners. Model-agnostic. Proxy-verified. Two devs enter, one leaderboard leaves.

A local proxy reads the provider's own usage field. No manual input. No guessing. No cheating.

0

Total Dome Throughput (tokens)

bolt connecting

—

Events/Min

hub

Providers

—

0

combatants

— events/min

workspace_premium

MVP of the Dome

@awaiting_champion

—

Personal Burn

trophy

Rank

Burner

Primary Engine

Total Tokens

Status

Deploy Your Proxy

Sign in with GitHub, grab your agent token, point your LLM clients at localhost. We handle the rest. Your prompts never leave your machine.

Enter the Dome →

key

House Rules

Counts From PROVIDER USAGE

Prompt Content STAYS LOCAL

Manual Input REJECTED

Local Models COUNT TOO

⚡ THE TOKENDOME

Verified Proxy Status

Two devs enter. One leaderboard leaves. Model-agnostic.

connecting…

Enter the Dome

Two steps. The first one puts you on the board in 60 seconds — no install, no code change.

Step 1 · Instant placement

Get on the board in 60s

Paste a read-only Admin key. We pull your last 90 days of usage — Claude Code, claude.ai, ChatGPT, direct API, everything on your org. No install. No code change.

rocket_launch

Includes Max-plan calls (Claude Code, claude.ai) plus any direct Anthropic API usage on the same org.

Get a key at console.anthropic.com → Settings → API Keys → Create Admin Key. Format: sk-ant-admin01-…

Admin key

⚠ Used once for this request. Never stored. Re-running overwrites the previous backfill (idempotent).

Days back

Step 2 · Stream live (optional)

Keep new tokens flowing in

The Admin-API import only catches what's already billed. To see new calls show up live as they happen, drop in the SDK shim or run the local proxy.

Your Agent Token — keep secret

—

Drop-in replacement for @anthropic-ai/sdk. Two lines.

1 — Install (token baked in)

npm i …/tokendome-anthropic.tgz

2 — Change this one line in your code

import Anthropic from '@tokendome/anthropic';

your x-api-key, prompts, and completions never leave your process

Drop-in replacement for openai. Two lines. Auto-enables stream_options.include_usage on streaming calls.

1 — Install (token baked in)

npm i …/tokendome-openai.tgz

2 — Change this one line in your code

import OpenAI from '@tokendome/openai';

One line installs the agent, logs you in, and registers it as a service that auto-starts on every login. Works for any language, any provider, any local model.

Paste in your terminal

click refresh to mint a 5-minute install code

The install code is short-lived and single-use, so your long-lived agent token never lands in shell history.

Then point any of your services at localhost:4000:

export OPENAI_BASE_URL="http://localhost:4000/v1"

export ANTHROPIC_BASE_URL="http://localhost:4000"

export OLLAMA_HOST="http://localhost:4000"

tokendome — bash — 80x24

$ tokendome login <your-token>

✓ logged in

$ tokendome start

Proxy: ACTIVE on :4000

Reporting: ENABLED

# fire up Claude / GPT / Ollama normally

_

Live

Scoreboard

Telemetry updated every 3s · provider-verified counts

Rank

Combatant

Total Tokens

Providers

Burn Share

Arena Throughput

— tok total

trending_up awaiting telemetry

Combatants

—

signed up and ready to burn

trophy

Current Champion

—

first to claim the dome takes the crown

Identity

Anonymize

Pick a public alias. Once set, your GitHub login and avatar are never displayed on the leaderboard, in profile drawers, in trash-talk attribution, or at the /u/<name> URL.

2–32 chars · letters, digits, . _ - · case-insensitive unique. Leave blank to revert to your GitHub login.

Public name

Ghost mode Hide me from the leaderboard, profiles, and trash-talk. My usage is still tracked privately — I just don't appear publicly.

Weekly digest email — optional, leave blank to unsubscribe

Mondays at 9am Pacific: this week's burn vs. last, top models, who passed you, any trash talk you got.

Agent token

If you accidentally shared your token (in a screenshot, screen-share, or repo), rotate it. The old token stops working immediately and you'll need to re-login each agent / SDK with the new one.