Docs
Hosted runs, stated once.
Everything about allowances, caps, top-ups and resets lives on this page. If a number is not here, it is not a rule.
What a run is
One run is up to 150,000 tokens of model work, counted across every agent, retry and tool call in a mission. It is an allowance unit, not a hard limit on a single task: a mission that goes past it spends another run.
Typical missions give you a feel for it. A scheduled check (open a page, read it, write a short summary) spends well under one run. An ordinary two-agent coding task spends one or two. A five-agent afternoon spends three to five.
Which models you get
Hosted runs use three pinned models: DeepSeek-V4.1-Flash-1017 by default, with GLM-5.3-Flash and Qwen3.8-Flash as fallbacks. The app chooses between them, including when a provider has an outage.
There is no model picker on a hosted plan. That is deliberate: the pinned default is both cheaper and stronger than the tiers it replaced, and one menu keeps the price predictable. If you want a specific premium model, bring your own API key — that path is free and stays free.
Allowance, cap and reset
Solo includes 180 runs a month, capped at 8 a day. Studio includes 500 runs a month, capped at 20 a day. The trial is 25 runs over 7 days, capped at 5 a day and requiring no card.
The allowance resets on your plan date and is shown as a date, never a countdown. Unused plan runs do not accumulate. The daily cap exists so a month cannot be spent in two evenings; at a steady 8 runs a day Solo lasts about 23 days and Studio about 25.
Top-ups
When the allowance is gone you can buy a prepaid pack: 60 runs for $6. Top-ups never expire, they are not subject to the daily cap, and they are spent only after the plan allowance is used.
Nothing auto-recharges. There is no overage invoice and no surprise charge; a mission that has no runs left is simply blocked with a card that offers the top-up.
What happens at zero
A mission already running finishes. New missions are blocked. The card shows your reset date and the top-up option. If you cancel, you keep the plan until the end of the paid month and then continue on your own key.
Your own key
Minotauris works without a plan. Add an API key from a provider you already pay and the app calls that provider directly: no runs, no allowance, no daily cap. Hosted runs are for people who would rather not manage provider billing and rate limits themselves.
What the server sees
A hosted request carries the mission’s model messages and tool definitions to the gateway, which forwards them to the selected provider. The gateway stores the run it charged, the model name and the time; it does not store your prompts or the model’s answers.
Your provider key is never needed for hosted runs, so there is no key of yours to leak. The hosted keys live only in the deployment environment.