You're a CTO / VP Eng / founder with cloud credits burning down. Checking on them is four logins and a 2FA prompt away — every single time — and when you finally get there, the console lies to you anyway (credit-funded accounts read $0.00 in the obvious places). So nobody checks. Tracking this stuff is a pain in the ass, and the number at the end of that pain is probably going to hurt 😉 — the least we can do is make it painless to look at. runway pulls live credit + spend facts from the cloud billing APIs and puts them one question away from wherever you already are:

Facts only. No forecasts, no traffic lights, no "insights". Your agent reads the same numbers you do and only speaks up when something genuinely warrants it.

Quickstart (5 minutes for AWS)

git clone https://github.com/ozkilim/runway.git
cd runway
./install.sh

Then follow docs/setup/aws.md — create a read-only IAM user (minimal policy JSON included, copy-paste), read your credit balance off the console once, fill in ~/.runway/config.yaml, and:

.venv/bin/python runway.py --check   # credentials smoke test
.venv/bin/python runway.py           # your numbers

If you use Claude Code, install.sh already installed the skill — just ask "what's my burn?". Any other agent (Cursor, Codex, Windsurf, ...): paste skill/AGENTS.md into your agent's instructions.

Providers

Provider Status Notes
AWS ✅ battle-tested Cost Explorer, true credit drawdown via record types
Azure ⚠️ partial Works for pay-as-you-go / EA. Blind for sponsorship offers — their usage never reaches the API; runway refuses to guess (why)
GCP 🧪 needs testers BigQuery billing export, true credit records. Wired, unverified against a real account — be the first
Yours? 🙌 ~150 lines to add one — see CONTRIBUTING.md

Design principles

These are what make the numbers trustworthy — hold new providers to them:

  1. Facts only. The tool emits what happened. Interpretation, forecasting, and panic are the reader's job (or their agent's, sparingly).
  2. UNKNOWN beats wrong. If a number can't be measured honestly, it renders as UNKNOWN with the reason attached — never a confident $0.00.
  3. Credits are first-class. Credit-funded accounts are the norm for startups, and every cloud console handles them badly. runway is explicit about granted / used / remaining / cash-beyond-credits.
  4. One manual read, then automation. No cloud exposes remaining credit via API. You read the grant off the console once (credits, credits_as_of, expires); runway measures drawdown from that anchor forever after.
  5. Secrets never leave. Read-only credentials, config at ~/.runway/ (never in the repo), secrets scrubbed from all error output.
  6. One provider failing never takes down the others.

CLI

runway.py            formatted table (what humans read)
runway.py --json     raw payload (what agents compute from)
runway.py --check    credentials smoke test, no cost data
runway.py --provider aws|azure|gcp
runway.py --config /path/to/config.yaml

FAQ

Why not just open the billing console? Four logins, a 2FA prompt, and three console menus deep — every single time. You will not. That's the whole point.

Why is credits_as_of so important? Remaining = credits − drawdown since that date. The console told you a balance on a specific day; measure from any other day and the answer is silently wrong. See the config comments.

Does this cost anything to run? AWS Cost Explorer API: ~$0.05/run. Azure/GCP: effectively free.

Is it safe? The credentials you create are read-only cost aggregates — they can't see resources or data, and can't spend money. Blast radius is "someone learns your bill".

Contributing

Adding a provider is the most valuable contribution — Oracle, Hetzner, DigitalOcean, Cloudflare, Vercel, Modal, RunPod, Lambda... anywhere burn hides. See CONTRIBUTING.md: one class, two methods, ~150 lines, with providers/aws.py as the reference.

MIT licensed.