Quota and limits

Replay Vision usage is metered per organization as estimated spend, on a monthly billing cycle. This page covers how observations are priced, what the budget is, what counts against it, and what to do when you're approaching it.

How usage is priced

Every successful observation has a fixed price that depends on the model the scanner uses – newer models tend to produce higher-quality observations but cost more per observation. The model picker in the scanner editor shows each option's price, and the current lineup is:

Prices are frozen when the observation runs: editing a scanner's model later doesn't change what past observations cost, and each $recording_observed event carries a credits property (100 credits = $1) with its price.

The monthly budget

Each organization gets a monthly spend budget – during the closed beta this defaults to $150 per month. Current usage and the projection for the rest of the period are shown on the Spend this period tile on the Replay Vision home.

You can also fetch the remaining budget programmatically – see the vision-quota-retrieve MCP tool.

What counts against the budget

Pending and running observations reserve budget so a burst of concurrent on-demand triggers can't blow past the cap before any of them complete. Once an observation reaches a terminal failed or ineligible state, its reservation is released.

Testing a configuration recommendation on the quality tab re-runs the scanner against rated sessions, and each tested session is charged like a normal observation of the scanner's model. The test controls show the planned cost before you run it.

What happens when the budget is reached

Both enforcement paths use the same check:

• Background sweeps – new observations queued by the automatic schedule are silently skipped. The scanner stays enabled, the sweep keeps running, but no work gets queued until budget frees up. A banner on the Replay Vision pages shows "Monthly spend limit reached" with the reset date.
• On-demand triggers – the player action and the vision-scanners-scan-session MCP tool return a 402 Payment Required response.

A scanner that's been paused by the budget resumes producing observations automatically once the period resets.

Tracking where spend goes

The Replay Vision home has a Usage tab with:

• Spend over time – a chart of your spend (daily, weekly, monthly, or yearly), with a forecast warning when the current pace would hit the budget before the period ends.
• A per-scanner breakdown – each scanner's successful observations this period, its price per observation (which model it uses), its sampling rate, and its share of total spend.

A scanner dominating the spend table is the first place to look when the budget runs down faster than expected – its sampling rate is the main cost lever.

Sizing scanners against the budget

The scanner editor shows an estimated cost for the scanner you're configuring – roughly how many observations per month and what they'll cost – based on:

• Your project's recent recording volume.
• The filters you've set.
• The session coverage mode.
• The sampling rate.
• The selected model's price.

If the projection is uncomfortably close to your remaining budget, the cheapest knobs are:

• Narrow the filters – e.g. require a specific URL pattern, exclude bots, restrict to a cohort.
• Tighten session coverage – switching from "All recordings" to "Skip lowest activity" or "Highest activity only" filters out low-activity sessions before sampling, so your budget goes toward recordings with meaningful activity.
• Lower the sampling rate – samples are random, so a 10% rate gives you a representative slice of matching sessions for one-tenth the cost.
• Pick a cheaper model – the price difference per observation is up to 7x across the lineup.

To project a hypothetical scanner before creating it (e.g. from a script or agent), use the vision-scanners-estimate-create MCP tool. It runs the same projection the editor shows.

What to do if you're hitting the budget

Try these, in order:

1. Open the Usage tab and look at which scanner is consuming the most spend. A scanner dominating the share-of-spend column is probably broader (or on a pricier model) than it needs to be.
2. Tighten its filters, session coverage, or sampling rate, or move it to a cheaper model. Use the editor's estimate to find a setting that gives you enough signal without dominating the budget.
3. Disable scanners you're not actively using. A disabled scanner produces zero new background observations.
4. Wait for the period to reset. The reset date is shown on the spend tile and in the budget banner.