gr8r cli · alpha

Your agent hits the limit.Your task doesn't die.

gr8r drives your coding agent. It reads how much of your subscription is left — the number your provider reports, not a guess — and when you're about to run out it writes down what's been done, wakes another model with it, and the task keeps going.

install

npm i -g https://getgr8r.com/gr8r-0.1.1.tgz

clipboard blocked by the browser — the line is selected, press ⌘C / ctrl+C

sha256 2966b84b0517e5569e85256cfa6b6a33560001881f5165a9c2313e91a51f56f5

Node 20+. Then run gr8r hull to see what's available on your machine.

before it can work

01 A main wind — the agent that does the work. Claude Code, signed in with your own subscription. No API key.

npm i -g @anthropic-ai/claude-code

clipboard blocked by the browser — the line is selected, press ⌘C / ctrl+C

02 A backup wind — where the task goes when the quota runs out. Anything speaking the OpenAI protocol. Free and local is fine.

ollama serve && ollama pull qwen2.5:3b

clipboard blocked by the browser — the line is selected, press ⌘C / ctrl+C

Without a backup wind there is nothing to swap to — gr8r will tell you so instead of pretending.

what it looks like

real run · 2026-08-01 claude-opus-4-8 → qwen2.5:3b · not a mockup

$ gr8r run --at=0.85 "add a test file next to each .js in src/"

  gr8r run · up to 4 legs
  wind    anthropic/claude-opus-4-8 (subscription)
  backup  local/qwen2.5:3b

  ── leg 1 · anthropic/claude-opus-4-8 ──
  ▸ Read src/math.js
  ▸ Write src/math.test.js
  ▸ Bash node --test src/
  seam: 13.0% of quota left · resets in 20h
  ⚡ wall at the seam — crossed 85.0% — provider says 87.0% used
     10 step(s) already done stay done — the work is on disk
     digest written · 22 lines · the thread is safe
     coupling local/qwen2.5:3b
  ── leg 2 · local/qwen2.5:3b ──
  ✔ task complete · 2 leg(s)
  ✔ the fuel ran out mid-task. The task did not die.

commands

gr8r run "<task>" Run the task. At the wall, swap wind and keep going. --at=0.9 swaps at 90% of quota instead of dying at 100%.

gr8r hull What's coupled, and what this machine actually has. gr8r hull backup ollama/qwen2.5:3b to couple one.

gr8r login Sign in with GitHub — a code, no password. The first task runs without it; from the second on, gr8r asks, because that is what keeps your history. gr8r whoami · gr8r logout.

gr8r status Who you are, coupled winds, and what this machine has learned about your ceiling.

gr8r demo A reproducible scene of the swap. No real model is called.

alpha

This is an alpha, and it is honest about it.

Tested on macOS only. Windows and Linux are unexplored — if you're there, you're the most valuable tester we have. Nothing is uploaded anywhere: everything gr8r knows lives in ~/.gr8r/ on your machine.