Jev AI Integrations: 6 Ways to Add Jev to Apps

What Changed for Jev AI Integrations?

The newest public update worth tracking is Vercel's September 21 guide to Jev AI integrations. It organizes the practical paths into AI SDK, TanStack AI, Cloudflare, LangChain, eve, and TypeSafe's SDK or HTTP API. That matters because Jev is no longer only a model announcement; developers now have several documented ways to call it from real application stacks.

The broader momentum is also visible in Vercel's AI Gateway launch data. Vercel reported that within 24 hours of launching on AI Gateway, Jev reached more paid teams than any prior model launch in Gateway history, with nearly 13% of paid teams using it by hour 24. That does not prove long-term adoption, but it is a strong signal that structured decision models are getting tested quickly in production environments.

Use this page as a routing guide. It does not claim that every integration has been tested by Jev AI Guides. Instead, it maps each documented route to the kind of application that should start there.

Choose the Right Jev AI Integration Path

Vercel's integration guide frames the choice around your existing application stack. That is the right starting point. A Jev integration should reduce glue code around structured decisions; it should not force a team to rewrite a working application around a new library.

The important distinction is between the library interface and the service receiving the request. AI SDK or TanStack AI defines how your code asks a question and reads the answer. Vercel AI Gateway, Cloudflare AI, OpenRouter, or TypeSafe receives the actual request. Mixing those layers up leads to wrong model IDs, wrong answer fields, and fragile migration work.

AI SDK and Vercel AI Gateway

AI SDK is the most natural route when Jev sits beside model calls in a TypeScript application. Vercel's guide says AI SDK uses experimental_evaluate for evaluation tasks and can send a Gateway model ID through Vercel AI Gateway.

The documented Vercel example uses typesafe-ai/jev as the evaluation model and asks Jev to classify an invoice-related support message into a team such as billing, account, or other. The output is a typed answer that application code can map to a support queue.

This route is especially useful for Next.js and Vercel teams that already use AI SDK. Keep the Jev call on the server. Do not expose Gateway or TypeSafe credentials in browser code.

TanStack AI Adapters

TanStack AI is a fit when your application already expresses decisions through decide() and adapter configuration. Vercel's guide lists a direct TypeSafe adapter, a Vercel Gateway adapter, a Cloudflare adapter, and an OpenRouter adapter.

The practical benefit is portability. You can keep the same state and decision question while changing the adapter that reaches Jev. That does not mean every field is identical. TanStack AI may expose a selected choice as value, while AI SDK examples use choice, and native TypeSafe uses its own response structure.

Use TanStack AI when adapter switching is part of the architecture. If you only need a simple API call, a direct SDK or HTTP route may be easier to maintain.

Cloudflare Workers

Cloudflare is the clearest route for teams already running code in Workers. The model identifier documented by Cloudflare is typesafe/jev, and the call uses env.AI.run().

This path keeps Jev close to edge or serverless routing logic. A Worker can evaluate a support message, form submission, moderation event, or risk signal, then pass a typed result into ordinary application logic. Cloudflare's Jev documentation also shows an HTTP route for account API access when a binding is not the right fit.

The main caveat is portability. The Cloudflare model ID and request wrapper should not be copied into Vercel or TypeSafe code without checking the target platform's own docs.

LangChain, eve, and Agent Workflows

LangChain is useful when the Jev decision belongs inside an existing Python workflow. Vercel's guide points to TypeSafeClassifier.invoke() and notes that LangChain exposes Choice, Score, and Noul question classes. This route is suitable for agent routing, tool checks, and workflow classifiers where LangChain already owns the orchestration.

eve is more agent-native. Vercel's integration guide describes Jev use inside evaluation, model-selection, and tool-approval helpers. In that shape, Jev does not write the agent's reply. It decides whether a proposed tool call is clear, cautionary, or needs a person; a separate language model still produces user-facing prose.

The repeated pattern is simple: Jev supplies a structured signal, but the application still owns the action.

TypeSafe SDKs, HTTP API, and Vercel Connect

For applications that do not use AI SDK, TanStack AI, LangChain, or eve, TypeSafe's SDKs and HTTP API remain the direct path. Vercel's guide references TypeSafe Python and JavaScript or TypeScript clients, plus POST https://api.typesafe.ai/v1/systemone for HTTP integrations.

Vercel also published a Jev connector page for Vercel Connect. That page focuses on credential handling: install the Connect skill, create a Jev connector, and request credentials at runtime through @vercel/connect. The point is not a new model capability. It is a safer way for apps and agents to access Jev credentials without copying secrets into application code.

The Vercel changelog also notes TypeSafe-compatible clients and an HTTP API for Jev on AI Gateway. That makes Vercel more than an AI SDK path: it can also be a routing layer for teams that want Gateway access while preserving a TypeSafe-style client shape.

Migration Checklist

When changing Jev AI integrations, keep the decision contract stable first. Then adapt the transport layer.

Jev accepts text-based state and returns typed decisions. Connecting it through a framework that also supports text generation, images, or agents does not turn Jev into a generator. Use it for classification, scoring, routing, and approval signals; keep generative models responsible for writing.

FAQ

What is the easiest Jev AI integration?

The easiest route depends on your stack. AI SDK is natural for Vercel and TypeScript apps, Cloudflare Workers can call env.AI.run('typesafe/jev', ...), and Python LangChain workflows can start with TypeSafeClassifier.

Can I use Jev through Vercel without AI SDK?

Yes. Vercel's September 21 changelog says AI Gateway supports TypeSafe-compatible clients and an HTTP API for Jev. The Vercel integration guide also describes a TypeSafe-compatible Gateway base URL and HTTP evaluation endpoint.

Is typesafe-ai/jev the same as typesafe/jev?

No. Vercel's guide uses typesafe-ai/jev for AI Gateway, while Cloudflare documents typesafe/jev for Workers AI. Copy the identifier for the platform you are actually using.

Should Jev control production actions by itself?

No. Jev should return a structured decision signal. Code should enforce permissions, thresholds, logging, fallbacks, and human review for consequential actions such as refunds, account changes, publishing, or command execution.