GatewayAgent Frameworks

Eve (gateway)

Eve accepts provider-authored AI SDK language models. Configure an OpenAI-compatible provider for Respan to route Eve model calls through the Respan gateway with one API key.

An Eve string model such as "openai/gpt-5.4-mini" uses Vercel AI Gateway. To use Respan Gateway, pass the provider model object shown below. AI_GATEWAY_API_KEY is a Vercel AI Gateway credential and is not used by this setup.

Setup

1

Install packages

$npm install eve@0.26.1 ai@7.0.26 @ai-sdk/openai@4.0.16

Eve 0.26 requires Node.js 24 or newer.

2

Set the Respan API key

$export RESPAN_API_KEY="YOUR_RESPAN_API_KEY"

No provider API key is required. The Respan gateway handles provider authentication.

3

Configure the Eve model

Create or update agent/agent.ts:

agent/agent.ts
1import { createOpenAI } from "@ai-sdk/openai";
2import { defineAgent } from "eve";
3
4const respan = createOpenAI({
5  apiKey: process.env.RESPAN_API_KEY!,
6  baseURL: "https://api.respan.ai/api",
7});
8
9export default defineAgent({
10  model: respan.chat("gpt-5.5"),
11});

With AI SDK 7, call respan.chat(...) explicitly. Calling respan(...) selects the Responses API, while this key-only configuration uses Respan’s Chat Completions route.

4

Run Eve

$npx eve dev

Switch models

Keep the provider configuration and replace gpt-5.5 with another identifier from the full model list. Provider prefixes make cross-provider routing explicit.

Eve’s provider-managed web_search tool is not available through this OpenAI-compatible Chat Completions route. Use an authored Eve tool when your agent needs web search.