Getting started

Call models from an action. The gateway authenticates with a short-lived token from getServiceToken("ai-gateway"). Model names use the provider/model form in Models.

OpenAI SDK

The gateway is OpenAI-compatible. Set baseURL to the gateway and pass getServiceToken as the API key:

npm install openai

import { action } from "./_generated/server";
import { v } from "bijection/values";
import OpenAI from "openai";
import { getServiceToken } from "bijection/server";

export const chat = action({
  args: { prompt: v.string() },
  handler: async (ctx, { prompt }) => {
    const openai = new OpenAI({
      baseURL: "https://ai-gateway.bijection.com/v1",
      apiKey: () => getServiceToken("ai-gateway"),
    });
    const completion = await openai.chat.completions.create({
      model: "openai/gpt-4o-mini",
      messages: [{ role: "user", content: prompt }],
    });
    return completion.choices[0].message.content;
  },
});

Agent component

Pass bijectionGateway as languageModel on @bijection-dev/agent. It obtains the token for you:

npm install @bijection-dev/agent @bijection/ai-sdk-provider

import { Agent } from "@bijection-dev/agent";
import { bijectionGateway } from "@bijection/ai-sdk-provider";
import { components } from "./_generated/api";

const agent = new Agent(components.agent, {
  name: "Support agent",
  languageModel: bijectionGateway("openai/gpt-4o-mini"),
  instructions: "You answer questions about our product.",
});

Workflows that call agent.generateText use that same model. For the RAG component, use bijectionGateway.embeddingModel(...) as its embedding model.

Vercel AI SDK

bijectionGateway works with generateText, streamText, embed, and embedMany from the Vercel AI SDK. Use @bijection/ai-sdk-provider with AI SDK 7.0.105 or later. The provider works in the default runtime and Node.js actions. For a Node.js action, set node.nodeVersion to "22" or "24" in bijection.json. The default Bijection runtime doesn’t need this setting.

npm install @bijection/ai-sdk-provider ai

import { action } from "./_generated/server";
import { v } from "bijection/values";
import { generateText } from "ai";
import { bijectionGateway } from "@bijection/ai-sdk-provider";

export const chat = action({
  args: { prompt: v.string() },
  handler: async (ctx, { prompt }) => {
    const { text } = await generateText({
      model: bijectionGateway("openai/gpt-4o-mini"),
      prompt,
    });
    return text;
  },
});

Stream with streamText:

import { streamText } from "ai";
import { bijectionGateway } from "@bijection/ai-sdk-provider";

const result = streamText({
  model: bijectionGateway("openai/gpt-4o-mini"),
  prompt,
});
for await (const chunk of result.textStream) {
  console.log(chunk);
}

Choose a model interface

For text generation, start with bijectionGateway(model), which uses Chat Completions across model providers. Use the native Messages or Responses interface when you need endpoint-specific provider features:

const messagesModel = bijectionGateway.messages("anthropic/claude-sonnet-4.5");
const responsesModel = bijectionGateway.responses("openai/gpt-5");

The Responses endpoint is stateless. The provider sets store: false; it does not support store: true or previous_response_id.

Decisions with Jev

Decisions is in alphabijectionGateway.evaluationModel() uses /alpha/decisions. The endpoint and AI SDK’s experimental evaluation interface may change during alpha.

Use AI SDK’s evaluate to classify or score data with Jev from an action:

import { experimental_evaluate as evaluate } from "ai";
import { bijectionGateway } from "@bijection/ai-sdk-provider";

const decision = await evaluate({
  model: bijectionGateway.evaluationModel("typesafe/jev-1.13"),
  state: { ticket: "Customer cannot sign in" },
  questions: {
    priority: {
      type: "choice",
      instructions: "Choose the response priority",
      criteria: {
        urgent: "Respond now",
        normal: "Respond today",
      },
    },
  },
});

console.log(decision.answers.priority.choice);

Questions can use choice, score, or boolean. Boolean answers contain a probability between 0 and 1. The provider translates this to the HTTP API’s noul question type and authenticates with getServiceToken("ai-gateway"). Pass abortSignal to evaluate to cancel a request.

Embeddings

Create embeddings with embedMany. The AI SDK automatically splits batches larger than the gateway’s 512-input limit:

import { embedMany } from "ai";
import { bijectionGateway } from "@bijection/ai-sdk-provider";

const { embeddings } = await embedMany({
  model: bijectionGateway.embeddingModel("openai/text-embedding-3-small"),
  values: ["hello", "world"],
});

Images and videos

Use bijectionGateway.imageModel() with generateImage, or bijectionGateway.videoModel() with experimental_generateVideo and experimental_startVideo. Image and video generation are in alpha. See Images and videos for examples and callback handling.

Manual fetch

Use fetch when you want the HTTP API without an SDK. Mint a token, then send it as Authorization: Bearer <token>:

import { action } from "./_generated/server";
import { v } from "bijection/values";
import { getServiceToken } from "bijection/server";

export const chat = action({
  args: { prompt: v.string() },
  handler: async (ctx, { prompt }) => {
    const token = await getServiceToken("ai-gateway");
    const response = await fetch(
      "https://ai-gateway.bijection.com/v1/chat/completions",
      {
        method: "POST",
        headers: {
          Authorization: `Bearer ${token}`,
          "Content-Type": "application/json",
        },
        body: JSON.stringify({
          model: "openai/gpt-4o-mini",
          messages: [{ role: "user", content: prompt }],
        }),
      },
    );
    return await response.json();
  },
});

For Decisions requests with fetch, see the HTTP API example. See HTTP API for request and response shapes.

Local development

Local development requires an up-to-date bijection package and local backend, and a deployment linked to a project whose team has AI Gateway access. From your project directory, run:

bijection login
bijection deployment select local
bijection dev

Restart bijection dev after signing in if it was already running, and accept the backend upgrade if prompted. The examples above, including getServiceToken("ai-gateway"), work unchanged. See Local deployments for more setup options. Inference runs in the cloud and is charged to the linked project’s team. Anonymous local deployments cannot use the gateway.

Token and timeouts

Call getServiceToken("ai-gateway") inside an action whenever a gateway request needs a credential. It returns a short-lived token scoped to your deployment; the action runtime caches and refreshes it as needed. Keep the token private. Don’t return it to clients, store it in environment variables, or cache it yourself. The request runs inside your action, so a long completion can hit the action timeout (30 minutes in the Bijection runtime, 10 minutes in Node) and fail as an action error rather than a gateway error. If getting a token fails with AiGatewayDisabled or AiGatewayUnavailable, see Who can use it.

Get Started

Platform

Data and Operations

Guides

Client Libraries

Tools

API Reference

OpenAI SDK

Agent component

Vercel AI SDK

Choose a model interface

Decisions with Jev

Embeddings

Images and videos

Manual fetch

Local development

Token and timeouts

​OpenAI SDK

​Agent component

​Vercel AI SDK

​Choose a model interface

​Decisions with Jev

​Embeddings

​Images and videos

​Manual fetch

​Local development

​Token and timeouts

OpenAI SDK

Agent component

Vercel AI SDK

Choose a model interface

Decisions with Jev

Embeddings

Images and videos

Manual fetch

Local development

Token and timeouts