How Cleo estimates your build

Every task Cleo's workers run uses AI processing, which has a cost. We translate this into a percentage of your monthly build capacity so you always know where you stand.

• Quick tasks (copy edits, small fixes, compliance checks) use very little — around 0.4% of a Pro monthly allocation.
• Standard builds (new feature, code review, testing run) use moderate capacity — typically 0.4–3%.
• Deep tasks (full MVP scaffold, architecture migrations, complex multi-file features) use the most — typically 3–15%.

We also weight usage by model — faster models like Haiku use 10× less allocation than Sonnet, so simple tasks cost proportionally less.

Software is inherently unpredictable. A feature that builds cleanly in one pass costs less than one that requires iteration and rework.

We show ranges rather than false-precision single numbers. All estimates include a 20% contingency buffer for unknowns — if your estimate says 10%, the real cost is unlikely to exceed 12%.

A small tooltip on every estimate chip explains this buffer. We never hide it.

• You'll see a warning at 80% and again at 95%.
• At 96%+, the allocation bar turns red with a prompt to upgrade or wait for your monthly reset.
• Allocation resets at the start of each calendar month — not on a rolling 30-day window.

We never silently charge you extra. You are always in control of when tasks run.

If you connect your own Anthropic API key, your tasks draw directly from your Anthropic account — not from the platform budget. No allocation bar applies.

Instead, you'll see a task count only: Using your own API key · 14 tasks this month.

You are only limited by your Anthropic account balance and rate limits.