Supported Embedding Models in OmniRoute: Complete Provider List (v3.8.51)

OmniRoute supports over 60 embedding models across 20+ providers including OpenAI, Gemini, Voyage AI, and Jina AI, all defined centrally in the open-sse/config/embeddingRegistry.ts configuration.

OmniRoute treats vector generation as a first-class capability, offering a unified API for cloud and local inference. All embedding models in OmniRoute are declared in the Embedding Registry, which maps provider IDs to model metadata including vector dimensions and display names. This guide enumerates every supported model as of release v3.8.51.

How OmniRoute Manages Embedding Models

The architecture relies on a single source of truth. In open-sse/config/embeddingRegistry.ts, the EMBEDDING_PROVIDERS constant declares every provider and its associated models. When a request hits the /v1/embeddings endpoint, the handler invokes getEmbeddingProvider() to resolve the model, or parseEmbeddingModel() to split a compound ID such as "openai/text-embedding-3-small" into its provider and model components.

The resolution flow follows these steps:

1. The API route src/app/api/v1/embeddings/route.ts receives the POST request.
2. The payload forwards to the embedding handler.
3. The handler calls getEmbeddingProvider() from the registry.
4. If valid, the handler builds an EmbeddingResolution (src/lib/memory/embedding/types.ts) and invokes the concrete implementation.

Because the registry is a simple JavaScript object, adding a model requires only updating this file—the rest of the stack automatically picks up the change.

Complete List of OmniRoute Embedding Models

The registry ships with 20+ providers. Below is the complete inventory organized by provider, including model IDs and vector dimensions where specified.

OpenAI and OpenAI-Compatible Providers

• OpenAI: text-embedding-3-small (1536 dimensions), text-embedding-3-large (3072 dimensions), text-embedding-ada-002 (1536 dimensions)
• Vercel AI Gateway: text-embedding-3-small (1536 dimensions), text-embedding-3-large (3072 dimensions)
• OpenRouter: openai/text-embedding-3-small, openai/text-embedding-3-large, plus native OpenRouter prefixes
• NanoGPT: text-embedding-3-small, text-embedding-3-large

Cohere

• embed-v4.0
• embed-multilingual-v3.0
• embed-multilingual-v3.0-images
• embed-multilingual-light-v3.0
• embed-multilingual-light-v3.0-images

Google Gemini

• gemini-embedding-001 (768 dimensions)
• gemini-embedding-2
• gemini-embedding-2-preview

OpenRouter also proxies these Gemini models under the google/ namespace.

Voyage AI

Most Voyage models output 1024 dimensions unless noted:

• voyage-4-large, voyage-4, voyage-4-lite
• voyage-3-large, voyage-3.5, voyage-3.5-lite (512 dimensions)
• voyage-multilingual-2
• voyage-code-3, voyage-code-2 (1536 dimensions)
• voyage-finance-2, voyage-law-2

Jina AI

• jina-embeddings-v5-text-nano (768 dimensions)
• jina-code-embeddings-1.5b (1536 dimensions)
• jina-code-embeddings-0.5b (896 dimensions)
• jina-colbert-v2 (128 dimensions)
• jina-embeddings-v5-text-small, jina-embeddings-v5-omni-small, jina-embeddings-v5-omni-nano, jina-embeddings-v4, jina-clip-v2

DeepInfra and Qwen

DeepInfra hosts multiple Qwen and BGE variants:

• Qwen/Qwen3-Embedding-8B (4096 dimensions)
• Qwen/Qwen3-Embedding-4B (2560 dimensions)
• Qwen/Qwen3-Embedding-0.6B (1024 dimensions)
• BAAI/bge-large-en-v1.5 (1024 dimensions)
• BAAI/bge-base-en-v1.5 (768 dimensions)
• BAAI/bge-m3 (1024 dimensions)
• intfloat/e5-large-v2 (1024 dimensions)
• thenlper/gte-large (1024 dimensions)

Mixbread, Nomic, and Specialized Providers

• Mixedbread: mixedbread-ai/mxbai-embed-large-v1, mixedbread-ai/mxbai-embed-2d-large-v1
• Nomic: nomic-embed-text (also available via Fireworks as nomic-ai/nomic-embed-text-v1.5 with 768 dimensions)
• NVIDIA: nvidia/nv-embedqa-e5-v5
• Upstage: embedding-query, embedding-passage (both 4096 dimensions)
• Mistral: mistral-embed (1024 dimensions)
• Together AI: BAAI/bge-large-en-v1.5 (1024 dimensions), togethercomputer/m2-bert-80M-8k-retrieval (768 dimensions)
• Nebius: Qwen/Qwen3-Embedding-8B (4096 dimensions)
• EmbeddingGemma: embeddinggemma
• BGE-M3: bge-m3 (available via multiple providers)

Local Model Placeholders

• LMStudio: Configurable for local models (no defaults listed)
• Ollama Local: Configurable for local Ollama deployments (no defaults listed)

How to Call OmniRoute Embedding Models

Use the generic endpoint for automatic provider resolution, or target a specific provider route.

Generic Endpoint

import fetch from "node-fetch";

const response = await fetch("http://localhost:20128/v1/embeddings", {
  method: "POST",
  headers: { 
    "Content-Type": "application/json", 
    "Authorization": "Bearer <API-KEY>" 
  },
  body: JSON.stringify({
    model: "text-embedding-3-small",
    input: ["Hello world!", "OmniRoute is awesome"]
  })
});

const data = await response.json();
console.log(data);

Provider-Specific Route

await fetch("http://localhost:20128/v1/providers/openai/embeddings", {
  method: "POST",
  headers: { "Authorization": "Bearer <API-KEY>" },
  body: JSON.stringify({
    model: "openai/text-embedding-3-large",
    input: "Embedding a single string"
  })
});

Programmatic Resolution

import { parseEmbeddingModel } from "@omniroute/open-sse/config/embeddingRegistry";

const { provider, model } = parseEmbeddingModel("openai/text-embedding-3-small");
console.log(provider); // "openai"
console.log(model);    // "text-embedding-3-small"

Key Implementation Files

Understanding these files helps customize the embedding layer:

• open-sse/config/embeddingRegistry.ts: Defines EMBEDDING_PROVIDERS and exports parseEmbeddingModel() and getEmbeddingProvider().
• src/app/api/v1/embeddings/route.ts: Handles POST requests to /v1/embeddings.
• src/app/api/v1/providers/[provider]/embeddings/route.ts: Handles provider-scoped routes such as /v1/providers/openai/embeddings.
• src/lib/memory/embedding/types.ts: Exports TypeScript interfaces including EmbeddingResolution and EmbeddingResult.
• src/shared/validation/schemas/apiV1.ts: Contains Zod schemas validating the embedding request payload.

Summary

• OmniRoute defines all embedding models in a central registry at open-sse/config/embeddingRegistry.ts.
• The platform supports 60+ models across 20+ providers including OpenAI, Cohere, Voyage AI, Gemini, and Jina AI.
• Vector dimensions range from 128 (Jina ColBERT v2) to 4096 (Qwen3 8B and Upstage).
• Local deployment placeholders exist for LMStudio and Ollama.
• Resolution occurs via parseEmbeddingModel() and validation uses Zod schemas in apiV1.ts.

Frequently Asked Questions

How do I add a custom embedding model to OmniRoute?

Edit open-sse/config/embeddingRegistry.ts and append a new entry to the models array within the appropriate provider object. Specify the id, name, and dimensions (if known). The API layer automatically recognizes the change without modifying route handlers or validation logic, as the registry serves as the single source of truth.

What are the vector dimensions for OmniRoute embedding models?

Dimensions vary by model architecture. Common sizes include 1536 (OpenAI small/Ada), 3072 (OpenAI large), 4096 (Qwen3 8B, Upstage), 1024 (Mistral, Voyage AI, BGE large), 768 (Gemini 001, BGE base, Nomic 1.5), and 128 (Jina ColBERT v2). The registry stores the dimension value in the dimensions field of each model descriptor.

Does OmniRoute support local embedding models via Ollama or LMStudio?

Yes. The registry includes placeholder providers for LMStudio and Ollama Local. While no default models are pre-configured, you can define local model IDs in embeddingRegistry.ts or configure the handlers to route requests to your local inference server endpoints.

How does OmniRoute validate embedding model requests?

Requests validate against a Zod schema defined in src/shared/validation/schemas/apiV1.ts. The handler first calls parseEmbeddingModel() to ensure the model string maps to a known provider in EMBEDDING_PROVIDERS. If the provider or model is missing, the API returns a validation error before attempting resolution.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}