Cycles

Appearance

Cycles Tools ​

Two interactive tools, one runtime-authority thesis. Cost calculators answer "how much will this workload spend?" Blast-radius calculators answer "if this workload's actions go wrong, how much damage is in scope?" Both questions matter in production. Both are bounded by Cycles at the same runtime gate.

Every configuration produces a unique URL. Share it, embed it, export it.

Available tools ​

Cost Calculator (Claude vs GPT) ​

Per-call, per-day, per-month, and per-year cost across current Claude and OpenAI models. Editable rates so you can plug in contracted pricing, add or remove rows for the models you actually use.

Blast Radius Risk Calculator ​

Monthly blast radius across agent action classes, modeled by reversibility (×1 / ×3 / ×10) and visibility (+0 / +1 / +4), with an editable runtime-authority containment factor. Highlights the catastrophic irreversible + public class with a red row outline. Name your agent so the artifact is self-attributed.

Try a pre-configured scenario ​

Each link below opens the calculator with realistic numbers from a specific incident or workload pattern, ready to tune to your own:

Each link encodes its scenario in the URL hash; you get the same view in any browser without a backend.

Share, embed, export ​

Each calculator's full state is encoded in the URL. Configure the calculator, click Share to copy a link with your numbers preserved, and the recipient sees exactly the same view.

The toolbar (visible in both the in-docs and full-screen views) offers:

  • Share — copies the standalone URL with current state
  • Copy — generates a Markdown table for Slack, GitHub PRs, email, or anywhere that renders Markdown
  • CSV — downloads a comma-separated file with attribution comment header
  • PNG — captures a branded screenshot of the table — runcycles logo, agent/workload name, source URL, and date are burned into the image
  • Embed — generates an <iframe> snippet to drop the calculator into a blog post, article, or vendor-comparison page. The embedded version preserves the URL state, so you publish a pre-configured calculator showing the numbers your post argues for.

Embed snippets look like this:

html
<iframe
  src="https://runcycles.io/calculators/claude-vs-gpt-cost-embed#s=..."
  width="100%" height="820" frameborder="0" loading="lazy"
  title="Cycles cost calculator"></iframe>

The embed view strips the docs chrome, shows only the calculator, and includes a small "Built with Cycles runtime authority — runcycles.io" credit linking back.

Why estimates do not equal runtime authority ​

These calculators help with capacity planning. They do not stop spend at runtime, and they say nothing about what an agent is permitted to do.

Estimating that a workload will cost $4,200/month is the start of the conversation. The rest is enforcing that it actually costs no more than $4,200/month — even when an agent loops, a tenant misuses your API, or a deploy regresses to a more expensive model — and that no single action (a refund, a deploy, a deletion) can cause damage that dwarfs your entire monthly LLM bill in one call.

If your projection includes the words "should not exceed" or "we expect," you are still relying on hope. See Why Cycles for the full runtime authority model — cost, action authority, blast radius, multi-tenant isolation — and Debugging sudden LLM cost spikes for the diagnostic playbook when reality exceeds the estimate.

Posts that use these calculators ​