Rystic runs fake Slack, GitHub and Stripe APIs on your machine, and the cheaper plans check in every 30 seconds

Anyone who has tested against someone else’s API knows the problem. The real service rate-limits you, charges you, or changes under you. A mock that returns canned answers lies in the other direction, because it doesn’t remember anything you did. Rystic is selling a middle path: a local copy of the API that keeps state. Its tagline for the idea is “Test against a simulator, not the live API.” The site’s explanation of why that works is a single sentence: “Every call sees the ones before it.”

So you create a Slack message, and a later call can read it. You can seed the starting state, control the clock, run scenarios, inject faults and latency, and look at the state in a UI at /_rystic/ui. A clock you can move is the underrated one. Testing “this invoice is 30 days overdue” against a real service means waiting a month.

The number that gets my attention is the match score. The site’s line: “Before every release, the same requests go to the live API and to the simulator. This is how often they agree.” The published figures: Slack 95.0% (840 of 884), GitHub 94.5% (587 of 621), PayPal 98.2%, Okta 98.6% and Kalshi 89.8%. On its fidelity page, Stripe is 97.2% (1,940 of 1,995). It’s good that the company publishes the counts and not just a percentage. I haven’t run any of it, so these are Rystic’s own tests of itself. And a simulator that agrees 95% of the time on requests chosen by its maker says little about the 5% that your code might hit first.

Pricing, from the pricing page: the Free plan is $0 for GitHub, Linear and Slack, one seat, 10 hours a month. Pro is $89.99 a seat a month with the full catalog, a $100 usage credit per seat, then $0.012 per simulator-minute, with a 14-day trial that takes a card. Enterprise is custom, with unlimited seats, and “fully offline.”

That last word matters because of what the cheaper plans do. According to the pricing page, Free and Pro simulators check in with lease.rystic.ai every 30 seconds, sending the API key and busy and idle windows. The page says it does not send request bodies. They allow one offline start within 24 hours and two hours after a lost connection. The page itself states it: “A fully offline machine needs Enterprise.”

I don’t see that as sinister. It’s how metered software works, and a $0.012-per-minute meter has to be read from somewhere. But it does change what “local” means. Your requests and your data stay on your machine, by the company’s account. Your usage doesn’t. If you test inside a network that blocks outside traffic, or on a plane, or on a client’s air-gapped box, the cheap plans won’t do the job for long.

For the right team this is a sound idea. Payments and chat APIs are the ones where the real thing costs money or sends messages to real people. A simulator that remembers is a safe place to break things. Just weigh it as three separate claims: it behaves like the real API (95% by its own count), it keeps your data local (by its account) and it works offline (only if you pay for Enterprise).

Offline still costs Enterprise money.