Honest credits — 1 credit = $0.001 of real frontier‑model compute, shown before you spend. Every run uses Opus 4.8, GPT‑5.5, or Sonnet 4.6 — no cheap models, no downgrades. Or bring your own key and skip credits entirely.
Use your own Anthropic or OpenAI key in the Builder — unlimited, free of our fees, key never leaves your browser.
Invite your team, share hosted agents, and bill per seat. Undercuts the competition — and you only pay for the seats you activate.
What counts as a run?
One blueprint generation or one sandbox/agent session. Storage of your agents is always free.
What does “bring your own key” mean?
Paste your own Anthropic or OpenAI API key and pay the model provider directly — we charge nothing on top, ever. Your key stays in your browser.
Do verdicts change for paying vendors?
Never. Our editorial firewall is structural: rankings and honest verdicts can't be bought.
Can I cancel anytime?
Yes — monthly plans have no lock-in. Your saved agents stay yours (and exportable) on any tier.
Is annual billing cheaper?
Yes — paying annually saves 20% versus monthly (about 2.4 months free). Switch the toggle above to see annual prices.
How do your credits work?
Honestly. 1 credit = $0.001 of real frontier-model compute — a published, fixed value. (Most tools' credits are an opaque unit worth anywhere from $0.0009 to $0.005, and you can't tell.) We show the cost before every run, let you set a hard spend cap, and never downgrade you to a cheap model. Or bring your own API key and skip credits entirely.
Which AI models do you use?
Only frontier models — Claude Opus 4.8 for blueprints, GPT-5.5 for interactive sandbox chats, and Claude Sonnet 4.6 for hosted automation (you can pin any agent up to Opus/GPT-5.5). No Haiku, no mini, no nano by default. The best model for the job, shown to you.
Prices in USD. Managed inference is metered with plan caps; hosted always-on automation is available on Plus and above.