three lanes · one wallet · live

Limitless tokens.
Any LLM. One wallet.

When Anthropic or OpenAI tells you to wait two hours, limitlesstokens.com doesn’t. Watch a 10-second ad, bank credits in cents, code past the rate-limit wall across three lanes — adtoken claude, adtoken codex, or adtoken open.

Limits are economic, not technical. We make them economic for someone else.

Install in 2 minutes Watch one ad
What one 4¢ ad watch funds · per lane
Lane 1 — Claude
~1.5 min
of Sonnet 4.6 coding
BYO-key forever ($9/M blended)
Lane 2 — Codex
~2.4 min
of GPT-5 coding
BYO-key forever ($5.63/M blended)
Lane 3 — Open
~17 min
of Llama 4 coding
centralized, no BYO-key ($0.80/M blended)
Lane 1/2 numbers reflect 2026 premium-frontier-model pricing. Lane 3 is the production lane — see loop economics for why the closed loop is provable today on open models. Sources: Anthropic · OpenAI · OpenRouter.

Install

Two minutes. Three commands.

# 1. Clone the repo
git clone https://github.com/henrytushman-ctrl/adtoken ~/work/adtoken
cd ~/work/adtoken && npm install

# 2. Paste any sk-ant-… (Anthropic) or sk-… (OpenAI) key — auto-detected.
#    AdToken mints your adt_ and writes ADTOKEN_API_KEY to your shell rc.
npx adtoken init

# 3. Watch ads to bank credits in your cents-denominated wallet.
adtoken earn 3

# 4. Spend on whichever lane you want — same wallet, three lanes:
adtoken claude "..."     # Lane 1 — Claude Code (BYO-key, Pro subscription untouched)
adtoken codex "..."      # Lane 2 — OpenAI Codex CLI (BYO-key, ChatGPT Plus/Pro untouched)
adtoken open "..."       # Lane 3 — Llama 4 / DeepSeek / Kimi / Qwen / GPT-OSS (no BYO-key)
adtoken resume          # auto-detects which lane your last session was on

One install, three lanes, one wallet. The operator sets per-lane revenue allocation on /admin; users watch ads and spend wherever they want.

What just happened

1
adtoken init auto-detects the provider from your key prefix
sk-ant-… wires up the Claude path; sk-… wires up the Codex path. Either way, AdToken mints your adt_<id> and writes ADTOKEN_API_KEY + ADTOKEN_USER_ID to your shell rc. Does NOT override the upstream API-key env var — your regular claude / codex keeps using subscription auth.
2
You code normally on your subscription (Lanes 1 + 2)
claude "..." and codex "..." work exactly as before. Pro/Plus rate limits, no per-token cost from AdToken, no AdToken in the picture.
3
You hit a rate limit. Anthropic / OpenAI say "wait."
Choices today: wait 2+ hours, upgrade ($200/mo Max or ChatGPT Pro), or stop coding.
4
Run adtoken claude "..." / adtoken codex "..." / adtoken open "..."
The wrapper sets the right BASE_URL + API_KEY env in the SUBPROCESS only. Lanes 1/2 route through your BYO key + AdToken's proxy. Lane 3 (open models) is fully ad-funded — operator runs an OpenRouter key, you spend banked cents. Shell env unchanged after the call.

Tools that work today

Anything that respects ANTHROPIC_BASE_URL, OPENAI_BASE_URL, or OPENROUTER_BASE_URL.

Claude Code Anthropic Python / TS SDK OpenAI Codex CLI OpenAI Python / TS SDK OpenRouter clients Aider Continue (VS Code, JetBrains) curl / direct HTTP

Doesn't work for:

Claude.ai web (OAuth) ChatGPT.com web (OAuth) Subscription-mode native CLIs (Claude Code Pro, Codex CLI Pro) Cursor's built-in Pro mode GitHub Copilot

These products use OAuth or proprietary billing rails — different lane than the API-credit market AdToken serves. (Per the No-Partnership Pivot, AdToken stays outside both Anthropic's and OpenAI's commercial billing forever.)

What we're going for

AdToken is not a chat UI. It is the credit layer underneath the tools developers already love.

You watch ads on this site once a day to top up credits. You code in your terminal the rest of the time. AdToken is invisible while you work.

Want to see the loop end-to-end first? Walk a demo session in the browser →