Limits and credits
Understand the current engine caps and the difference between monthly credits and per-scan limits.
Sight uses both per-scan engine caps and recurring plan allocations. Per-scan caps bound crawl depth, rendering, performance tests, GEO questions, model fanout, samples, judged pages, competitors, and internal model budget. Monthly credits govern how much recurring model work an organization can consume across scans.
These systems protect cost and runtime but need to remain visible to the person interpreting results. Before comparing projects or plans, confirm which work was eligible, what actually ran, when monthly usage resets, and how scheduled scans behave when the remaining allocation cannot cover the requested configuration.
Limit visibility is part of result quality. The workspace should make it clear when a cap was reached, which pages or prompts were included, and whether later work was skipped. Teams can then reduce scope, upgrade, or split an investigation intentionally. Without that context, partial evidence may be mistaken for a complete audit and apparent changes may reflect coverage rather than site behavior.
| Per scan | Free | One-time | Starter | Pro |
|---|---|---|---|---|
| Crawled pages | 5 | 100 | 100 | 250 |
| Browser-rendered pages | 1 | 10 | 10 | 25 |
| Performance-tested pages | 1 | 5 | 5 | 10 |
| GEO questions | 8 | 10 | 40 | 80 |
| Probe models | 2 | 4 | 4 | 5 |
| Samples per prompt and model | 1 | 1 | 2 | 3 |
| LLM-judged pages | 1 | 10 | 15 | 25 |
| Competitors scored | 0 | 2 | 2 | 3 |
| Internal LLM budget | €0.20 | €0.60 | €0.60 | €1.50 |
Monthly credits
Starter receives 10,000 credits and Pro receives 30,000 credits per UTC calendar month. Credits fund monitored scan model responses; onboarding assistance and generated chat titles are not billable.
Track allocation, consumption, remaining balance, and reset date together. A usage view should explain which project, scan, prompt, model, or sample consumed the amount. Customers need that ledger to adjust configuration before a recurring workflow reaches its limit.
| Model family | Credits per response | Ledger attribution |
|---|---|---|
| Google or Gemini | 12 | Project, scan, provider, model, hashed prompt reference, tokens, and cost |
| OpenAI or GPT | 18 | Project, scan, provider, model, hashed prompt reference, tokens, and cost |
| Anthropic or Claude | 24 | Project, scan, provider, model, hashed prompt reference, tokens, and cost |
| Perplexity or Sonar | 28 | Project, scan, provider, model, hashed prompt reference, tokens, and cost |
| Other configured model | 18 | Project, scan, provider, model, hashed prompt reference, tokens, and cost |
Reservation and settlement
Sight reserves credits before a billable provider request, records actual tokens and cost after success, and writes an idempotent refund adjustment after failure. Unused credits receive an immutable expiry entry at the next monthly boundary and do not roll over. Work that cannot reserve enough credits is skipped.