Setup

Install the adapter beside Anvia core. The official SDK it uses requires Node.js 20 or newer.

pnpm add @anvia/core @anvia/jev

Set TYPESAFE_API_KEY in the server environment and create the client in a server-only module:

import { JevClient, JEV_LATEST } from '@anvia/jev'

export const jev = new JevClient({
  apiKey: process.env.TYPESAFE_API_KEY,
})

export const decisionModel = jev.decisionModel({ modelId: JEV_LATEST })

apiKey falls back to TYPESAFE_API_KEY when omitted. The constructor throws a TypeError when no key is available or the key is blank. Keep keys and clients out of browser bundles.

Client options

JevClient accepts either managed options or an injected client:

• apiKey authenticates the SDK client the adapter creates.
• baseUrl points it at a different TypeSafe-compatible endpoint.
• headers adds default request headers.
• client reuses an existing TypeSafeClient.

const jev = new JevClient({
  apiKey: process.env.TYPESAFE_API_KEY,
  baseUrl: process.env.TYPESAFE_BASE_URL,
  headers: { 'X-Application': 'catalog' },
})

Keep endpoint and header values in trusted deployment configuration.

Bring an existing SDK client

import { TypeSafeClient } from '@typesafe-ai/sdk'

const sdk = new TypeSafeClient({ apiKey: process.env.TYPESAFE_API_KEY! })
const jev = new JevClient({ client: sdk })

Injection is mutually exclusive with apiKey, baseUrl, and headers; combining them throws. Managed clients disable SDK logging, while injected clients keep their own logging configuration. In both cases the adapter disables SDK retries per call, so decide({ retries }) controls retry behavior.

Select a model

modelId is required. JEV_LATEST ('jev-latest') provides autocomplete, and any other non-empty string is accepted. A blank ID throws a TypeError. Treat the ID as deployment configuration and test it against your account.

Continue with Decisions.