@voxell/forge-ai-provider

A Vercel AI SDK community provider for Forge — Voxell's hosted text-embedding API. Use Forge with embed / embedMany like any other AI SDK provider.

Why Forge

• Quality you can dial. Forge runs the Qwen3-Embedding family; ultra is the 8B — ~75+ average task score on MTEB, currently #4 on MTEB (English), and the top usable model (the three ranked above it are research-only). turbo (0.6B) is the fast/cheap tier — pick your point on the quality/cost curve.
• Matryoshka (MRL). Truncate to fewer dimensions (re-normalized) for ~4× smaller, cheaper vectors — one option, minimal quality loss.
• Low latency. Served by a Go + CUDA engine; HTTP and gRPC.
• Zero-trust. Per-key auth; mTLS available.
• Start free. 10M tokens, no card — dash.voxell.ai. Learn more: voxell.ai/forge.

Install

npm i @voxell/forge-ai-provider ai

Use

import { embed, embedMany } from "ai";
import { forge } from "@voxell/forge-ai-provider"; // API key from FORGE_API_KEY

const { embedding } = await embed({
  model: forge.textEmbeddingModel("turbo"),
  value: "hello world",
});

const { embeddings } = await embedMany({
  model: forge("pro"),            // shorthand for textEmbeddingModel
  values: ["doc one", "doc two"],
});

// Matryoshka — smaller, cheaper vectors (re-normalized server-side)
await embed({
  model: forge("turbo"),
  value: query,
  providerOptions: { forge: { dimensions: 256 } },
});

// Retrieval: mark queries vs documents
await embed({
  model: forge("turbo"),
  value: query,
  providerOptions: { forge: { inputType: "query" } }, // default is "document"
});

Models

| id | dimensions | notes | |----|-----------|-------| | turbo | 1024 | fast, low cost — default | | pro | 2560 | | | ultra | 4096 | Qwen3-Embedding-8B; ~75+ avg MTEB, #4 (English), top usable model |

Configuration

import { createForge } from "@voxell/forge-ai-provider";
const forge = createForge({
  apiKey: process.env.FORGE_API_KEY, // default
  baseURL: "https://api.voxell.ai",  // default
  headers: {},
  // fetch: customFetch,
});

providerOptions.forge: dimensions (Matryoshka truncation) · inputType ("query" | "document").

License

MIT © Voxell, Inc.