> ## Documentation Index
> Fetch the complete documentation index at: https://docs.errorbar.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Plans

> Free, Pro and Scale on top of prepaid usage. Plans buy the loop; usage stays at the rate card.

Every errorbar workspace pays for usage the same way: prepaid, per token, at the
published rate card. A **plan** sits on top of that and buys the things that
make the platform the product — the right to *act* on a verdict, longer
retention, priority on the eval executor, a monthly usage credit, and a larger
allowance of scored calls. A plan never changes a per-token price.

|                               | Free                          | Pro          | Scale          | Enterprise          |
| ----------------------------- | ----------------------------- | ------------ | -------------- | ------------------- |
| Price                         | \$0                           | \$99 / month | \$299 / month  | contract            |
| Usage credit                  | \$5 once, on a verified email | \$50 / month | \$150 / month  | contract            |
| Gateway rate limit            | 300 rpm                       | 900 rpm      | 3,000 rpm      | 6,000 rpm+          |
| Eval priority                 | 1×                            | 2×           | 4×             | 8×                  |
| Concurrent eval runs          | 1                             | 2            | 5              | 8                   |
| Request logs (opt-in content) | 14 days                       | 30 days      | 90 days        | 1 year              |
| Telemetry (content-free)      | 30 days                       | 90 days      | 1 year         | 2 years             |
| Scored calls included / month | 5,000                         | 50,000       | 250,000        | 1,000,000           |
| Workspaces                    | 1                             | 1            | 5              | unlimited           |
| Gates                         | recommend                     | auto-enforce | auto-enforce   | auto-enforce        |
| Alerts · alias canary %       | locked                        | ✓            | ✓              | ✓                   |
| Dedicated GPU endpoints       | —                             | —            | GPU-hour + 15% | GPU-hour + 10%, SLA |
| Dedicated judge endpoint      | —                             | —            | —              | ✓                   |

The numbers on this page are the numbers the platform enforces: the gateway
rate-limits by plan, the eval executor prioritises by plan, and the scored-call
meter counts against the plan's allowance.

## What "the verdict is free" means

On Free you can run every comparison, calibrate every judge, and see every gate
verdict, alert condition and canary option. What Free does not do is **act**:

* a gate can be in *recommend* mode (verdicts recorded), not *auto* (repoints by itself);
* alerts cannot be turned on;
* an alias cannot carry a canary percentage.

Each of those controls stays visible in the dashboard, disabled, with the one
sentence that opens it (`Upgrade to Pro to turn on alerts.`). The public API
refuses the same actions with **`402 upgrade_required`** — distinct from the
balance `402 insufficient_balance`, so an integration can tell "top up" from
"upgrade".

## Usage credit

Pro and Scale grant a usage credit with every paid invoice: $50 and $150. It is
spent **before** your prepaid balance on gateway tokens and scored calls, and
whatever is unused **expires at the end of the billing period** — it is an
allowance, not a top-up. Your wallet shows the credit remaining and when it
expires; the ledger marks how much of each charge came from credit.

Plan credit does not apply to dedicated GPU-hours, fine-tuning or batch jobs;
those draw on the prepaid balance.

## Scored calls

A **scored call** is one judge verdict on one thing: an eval judgment, an
alignment or calibration verdict, an online-monitoring score, a canary-gate
score, a suspect scan, a dataset judge-gate. Each plan includes an allowance per
calendar month; above it, every scored call adds **\$1 per 1,000** as its own
rate-card line — on top of the judge's own tokens, which are always billed at
the rate card. A verdict we had to re-parse is billed for its tokens but not
counted twice. The billing page shows the month's count against the allowance.

## Retention

The retention windows in the table are ceilings. Each workspace chooses its own
window for opt-in request logs and for telemetry; the plan caps the choice and
every stored row carries its own deadline. Moving to a shorter plan clamps the
windows and brings already-stored deadlines forward.

## Changing plans

Upgrade from **Settings → Billing** (Stripe Checkout). The plan applies the
moment the subscription is active; the credit lands when the invoice is paid.
Change or cancel from **Manage billing** (the Stripe portal). A cancelled plan
runs to the end of its period, then the workspace returns to Free, any
remaining plan credit is forfeited, and retention clamps to the Free ceilings.
Your prepaid balance is never touched by a plan change.

Enterprise is a conversation, not a card: [talk to us](https://www.errorbar.ai/contact).