Question 1

Is the TypeSafe adapter a real System One call with typed answers, or just routed chat completions?

Accepted Answer

It is a first-class adapter in the same OpenAIService as Anthropic, Google Gemini/Vertex, Groq, Perplexity including its decision model, Liquid AI d1, and the decision models on OpenRouter. The README does not spell out the Jev method signatures or which question types the adapter exposes, so those come from the adapter module rather than that page.

Question 2

Will I be pulling in a huge amount of API surface I do not need just to call Jev?

Accepted Answer

Yes, that is the trade. One service covers every OpenAI endpoint plus Anthropic including Bedrock and Managed Agents, Google Gemini/Vertex AI, Groq, Perplexity, TypeSafe AI, Liquid AI d1 and OpenRouter decision models. The README says the library was built self-contained with few dependencies, including its own generic WS client.

Question 3

Several endpoints in that list are marked deprecated. Will they still work?

Accepted Answer

Not all. createCompletion, createEdit, createImageVariation, createChatFunCompletion and createAudioTranslation are each flagged deprecated, and Assistants, Threads, Thread Messages, Runs and Run Steps are deprecated since 1.4.0 with the Responses API recommended instead. OpenAI-compatible servers keep createCompletion through OpenAICoreService.

Question 4

What exactly does it take to get the Jev path running in a Scala project?

Accepted Answer

Scala with sbt, an MIT licence, and ready-to-use examples in the openai-examples module. The README documents only the general shape, so the credentials and base URL a non-OpenAI adapter expects at construction time are not stated there; you would need to read the adapter source before wiring it up.

Question 5

Which providers sit behind the same service?

Accepted Answer

Anthropic first, including Bedrock and the Managed Agents surface of agents, environments and their work queue, sessions, deployments, vaults and credentials, and memory stores, plus native Messages with typed stream events, Message Batches, Files and Skills. Then Google Gemini and Vertex AI, Groq, Perplexity including its own decision model, TypeSafe AI for Jev, Liquid AI d1, Microsoft Foundry with Microsoft-Decision-1, and the decision models on OpenRouter. All of it is reached through one OpenAIService.

Question 6

How does streaming work across all those providers?

Accepted Answer

One provider-neutral typed stream of text, thinking, tool-call and tool-result chunks, including Anthropic's native Messages events, rather than a per-provider stream shape. Streaming is available on the normal endpoints, and the Responses API adds createModelResponseStreamed, which emits typed events or ChatChunks depending on which you ask for. Tool calling and MCP tools ride the same chunk types.

Question 7

What version is it, and how self-contained is the build?

Accepted Answer

Version 1.5.0, MIT licensed, with ready-to-use examples in the openai-examples module. The stated goal was a self-contained library with the fewest dependencies possible, including implementing its own generic WebSocket client rather than pulling one in. Service function names match the OpenAI API endpoint names exactly in camelCase, which is why the Jev entry point is easy to miss if you are grepping for something TypeSafe-shaped.

OpenAI Scala Client

About this project

Highlights

Questions about this entry

From the Jev guides

From the Jev AI model community