---
url: https://scruple.dev/providers/jev.md
---
# Jev provider

`@scruple/provider-jev` sends bounded code evidence and fixed questions to Jev through
`@typesafe-ai/sdk`. This includes final candidate decisions and any collection classifications a rule
uses to recognize ambiguous candidates. Requests use HTTPS. TypeSafe publishes
[current Jev model pricing](https://docs.typesafe.ai/models). Jev 1.13 is currently $0.042 per million
input tokens, with output tokens free.

```sh
pnpm add --save-dev @scruple/provider-jev
```

```ts
import { jevProvider } from "@scruple/provider-jev";

const apiKey = process.env["TYPESAFE_API_KEY"];
if (apiKey === undefined) throw new Error("TYPESAFE_API_KEY is required");

const provider = jevProvider({ apiKey });
```

## Options

| Option        | Default        | Purpose                                |
| ------------- | -------------- | -------------------------------------- |
| `apiKey`      | Required       | Non-empty Jev API key                  |
| `model`       | `"jev-1.13.0"` | Model sent with each request           |
| `baseURL`     | SDK default    | Alternate API endpoint                 |
| `concurrency` | `64`           | Maximum simultaneous provider requests |
| `timeoutMs`   | `10000`        | Request timeout in milliseconds        |
| `maxRetries`  | `2`            | SDK retry count                        |
| `fetch`       | Runtime fetch  | Custom fetch implementation            |

The provider reports input and output token usage in Scruple's run statistics. Uncached request and
token totals include collection classifications as well as final candidate decisions. The CLI reports
cached responses separately as `cacheHits`.

Review the Jev service's current data handling and billing terms before sending repository code. Scruple does not make claims here about service retention or model training.

## Performance

See the [benchmark results](../reference/benchmarks.md) for measured latency and throughput across concurrency levels. The current results support the default concurrency of `64` for sustained workloads.