Do we pay for runs on our own machines?
No usage fees, on any plan, and none on Enterprise hardware rigs either. Plans limit how many of your machines are connected at the same time. You only pay for usage when you choose Invariant cloud workers.
What counts as a connected runner?
A machine that is connected to Invariant and ready to take jobs, counted at the same moment. A laptop, a GPU server or a CI job each count as one while connected. Short CI runners count only while their job runs. If you hit the limit, extra runners wait to connect; we never upgrade your plan automatically.
Can we use our own CI and coding agents?
Yes, that's the intent. Invariant runs as a step in the CI you already use, and coding agents can call the CLI on any plan or the MCP server on Team and above. The spec format, CLI and local runner are planned to be open source.
Is pricing per seat?
No. Every plan, including Free, has unlimited users, so everyone who touches the robot can see results. Plans differ in automation, history and included cloud credit.
Which plan should we start on?
We recommend starting on Free with one log and one metric, move to Team when the test runs on every pull request, and to Startup when they want bisect, chat alerts and a year of history. You can change plans at any time.
What does a cloud worker cost in practice?
Workers are billed per second with a one-minute minimum. The pull request in our example, 210 paired runs on 8 CPU and 4 L4 workers for 7 minutes 41 seconds, comes to about $1.50. Each worker is billed separately.
What happens when the monthly credit runs out?
Runs continue at the listed rates up to a spend limit you set. When you reach it, cloud runs pause and your own runners keep going. Unused credit doesn't roll over.
What does one plan cover?
One company or lab. A plan covers everyone there, across all of its projects. If you need several separate organizations, SSO or your own infrastructure, that's Enterprise.
Can we start with a pilot?
Yes. We're looking for our first design partners. We set up one workflow with your team, usually log replay for localization or perception, and agree on pilot terms and a success measure before we start.
Do you use our data?
Your logs, models and outputs stay in your bucket. Invariant stores run records, metric values and verdicts, and never uses your data to train models.
What does "Coming" mean?
Invariant is in early access. The first release covers log replay on your own machines; features marked Coming are on the roadmap and are included in the plan when they ship, at no extra cost.