Priced on work done, not on headcount

A unit of analysis we run for you. A pull request reviewed is 10, an issue triaged is 5, a push indexed or a context captured is 1. Reading the graph over MCP is free and always will be, because querying what you already own should not be rationed.

A pure meter makes the bill unpredictable and punishes exactly the teams doing the most with agents. A band gives you a number you can budget, with overage only if you go past it. Most teams never do.

No. Overage is the platform meter: events past your included allowance, charged whether or not you bring your own keys. Model usage is separate and is only on your bill if we are running the models for you.

Because our costs behave differently. Platform cost per event falls with volume: fixed infrastructure amortises and storage is bought in bulk, so a bigger commitment earns a lower marginal rate. Inference does not work that way. We pay roughly provider list whatever the volume, so cloud models stay at cost plus 10 percent on every plan. Tiering that would mean inventing a discount we do not actually get.

Nothing stops. Events past the included allowance bill at the overage rate for your plan, and we tell you when you cross so it is never a surprise at the end of the month.

No. Every overage rate is set above what the next plan charges for the same volume, so by the time you would reach the next band, moving up already costs less. We would rather you paid us less than felt trapped on a plan you had outgrown, so we tell you when you cross the line.

Whatever band your event volume lands in, and nothing for the engineers. A monorepo counts as one repository, so a large team on few repositories usually sits lower than its headcount would suggest. Talk to us and we will size it from your real pull-request and commit volume rather than guessing.

Yes, including the cheapest one. Our buyers tend to have provider contracts, committed spend, or rules about which models may see their code, and none of that should force an upgrade. Connect a key and we bill you nothing for inference.

Yes. You are billed for the tokens actually consumed, and a run that had to be retried consumed them twice. It is itemised, so a retry is visible on the invoice rather than folded into a rounded-up total. Bringing your own keys does not change this either: your provider bills the same tokens, they just bill you directly.

What the provider charged us plus 10 percent, itemised on your invoice so you can check it against their pricing page. The margin is deliberately thin: you are already paying for the platform, and we would rather run the models as a convenience than charge you twice for the same month of work. No minimum and no commitment: switch to your own keys whenever you like and the charge stops.

Any tool that speaks MCP. Context and graph queries are served over a standard interface, so you are not locked to one assistant.

No. Every paid plan carries the whole managed layer. Plans differ only by how many repositories you hold and how much analysis you run, plus whether you bring your own model keys. What separates self-host from paid is the managed layer itself, not a feature flag inside it.

Yes, and it is a real product rather than a demo. Self-host the engine with no repository or event limits, run review and triage on your own servers with your own keys, and serve the graph over MCP. A paid plan adds the layer a team works in: curated memory, the panel, and the applications built on both.

Yes. The on-prem plan deploys the full cloud experience inside your network, premium plugins included, and supports air-gapped setups with your own model endpoints. There is no event metering on-prem. Talk to us at [email protected].