Form → AI (Gemini Flash-Lite) → Sheets → LinkedIn ≈ 0 + (10 + 36) + 10 + 20 = 76
Form → AI (Claude Sonnet 5) → Sheets → LinkedIn ≈ 0 + (10 + 175) + 10 + 20 = 215
Form → own-key AI → Sheets → LinkedIn = 0 + 10 + 10 + 20 = 40
Manual → Loop → 10 Sheets invocations = 0 + 0 + 10 × 10 = 100The AI examples assume a 1,500-token prompt and a 400-token answer. Administrators can change prices; the picker, the configuration panel and the public pricing page show the current ones. Each invocation keeps the prices it was admitted with even if they change later. Workflow estimates show a typical cost (from your recent runs where available) and an upper bound; neither is a guaranteed total.
| Node or operation | Credits |
|---|---|
| Workflow start, triggers, Agent configuration | Free |
| Logic and flow: Condition, Filter, Loop, Delay, Wait, Approval, Transform, Expression | Free |
| App actions: HTTP, MCP, Gmail, Sheets, Slack, Linear, WordPress, BigQuery, Apify, invoices | 10 per invocation |
| LinkedIn, LinkedIn draft, Instagram Reply with Content | 20 per invocation or interaction |
| AI / AI Agent / AI Caption on Superagento credentials | 10 step fee when it succeeds, plus model usage per token |
| AI / AI Agent / AI Caption with your own API key | 10 step fee; your provider bills the tokens |
Model usage is charged at the selected model's rate in credits per 1,000 input and output tokens (cheaper for cached input), as listed on the pricing page. Before each model call, the call's maximum is held from your balance; the unused part returns as soon as the provider reports usage. An Agent's tool calls and internal model calls are each metered this way, and one AI invocation is capped at 12,500 credits.
If an AI step fails after the model answered, you pay for the tokens it used but not the step fee. Errors returned by the provider, and failures before anything is sent, cost nothing. If the provider never answers, that call stays held until it is reconciled, and is never resent automatically. Each plan also stops a single run at a credit limit (1,000 on Free, 10,000 on Starter, 50,000 on Pro), so one runaway loop cannot spend a month's allowance.
Insufficient funds stop a node before any provider work; earlier completed nodes stay charged. Reserved credits are displayed separately from settled use. Model, token, tool and runtime limits still apply with your own key.