"How much does a run cost?" is the most common question after "how do I start?" Fair enough. Here's the full picture.
Credits, not dollars
Everything in Svortie runs on credits. You buy credits with money, spend credits on runs. One layer of abstraction, but it keeps pricing simple across different resources.
You get credits two ways:
Subscriptions give you a monthly credit allowance that rolls over:
- Starter: 1,500 credits/month for $9 ($90/year)
- Pro: 6,000 credits/month for $29 ($290/year)
- Power: 20,000 credits/month for $79 ($790/year)
Yearly plans save roughly two months.
Credit packs are one-time purchases if you need more:
- 500 credits for $5
- 2,000 credits for $20
- 5,000 credits for $50
What eats credits
Four things consume credits:
AI tokens are the big one. Every prompt and response costs tokens, and different models have different rates per 1,000 tokens. Haiku is the cheapest. Sonnet costs more. Opus costs the most. Your billing page has a "Credit Rates" section with exact per-model rates, calculated from current model pricing.
Sandbox compute charges per minute while your agent's environment is running. Usually a small fraction of the total.
Speech-to-text costs credits per minute of audio if you dictate tasks instead of typing.
Email notifications cost a small flat amount per send.
What runs actually cost
Real numbers from real runs:
A research agent (Haiku, web search, 30 iterations) typically costs 10-20 credits per run. That's a few cents.
A 3-step workflow (three Haiku agents) runs 30-60 credits total. Under a dollar.
A code review agent (Haiku, reading and analyzing a codebase) runs 15-40 credits depending on repo size.
The same tasks on Sonnet cost roughly 5-10x more. On Opus, significantly more than that. This is why we default to Haiku and recommend workflows of cheap specialists over single expensive generalists.
Reading the usage dashboard
Go to Usage. Four summary cards at the top: credit usage, agent runs, workflow runs, and scheduled runs. Click any card to filter the chart to that metric.
Credit usage breaks down as a stacked area: LLM, sandbox, speech-to-text, and email. Toggle between 7, 30, or 90 day views.
Low balance alerts
On your billing page, enable low balance alerts. Set a threshold (say 100 credits) and a frequency (every hour, every 6 hours, once a day, or once a week). Alerts trigger after a run completes with your balance below the threshold.
No surprises
Every completed run shows its credit cost right in the run detail. Before you run, you can estimate: check the model rate on your billing page, multiply by expected token usage. But honestly, Haiku runs are cheap enough that most people just run and check after.
What's next
- Create your first agent if you haven't already
- Build a workflow to get more done per credit
- Read the cost comparison for the case against expensive models