bedda.ai Spotlight: Embeddings API — Semantic Search and RAG for Developers

Feature spotlight: Embeddings

Every week we highlight one bedda.ai feature. This week: embeddings — the developer-facing API for generating vector embeddings across multiple models, unified under a single endpoint.

What embeddings are for

Embeddings are numerical representations of text that capture semantic meaning. They're the foundation of:

• Retrieval-Augmented Generation (RAG) — embedding your documents so an AI can retrieve relevant context before answering
• Semantic search — finding documents by meaning, not just keyword match
• Recommendation systems — finding similar items by embedding their descriptions
• AI memory — storing and retrieving conversation context or user preferences

Every production AI application that works with external documents or large knowledge bases uses embeddings.

The multi-provider problem

Like text AI models, embedding models vary significantly by provider and version. OpenAI's text-embedding-3-large produces 3072-dimensional vectors with high semantic precision. Cohere's embed-english-v3 is optimized for retrieval tasks. Open-source alternatives like bge-large offer competitive performance at lower cost.

The problem: each provider has a different API, different authentication, different response schema, and different token limits. Experimenting with or switching embedding models means rewriting API integration code.

bedda.ai Embeddings API

bedda.ai exposes a unified embeddings endpoint:

// Same call — switch the model parameter to change providers
const response = await fetch('https://bedda.ai/api/embeddings', {
  method: 'POST',
  headers: {
    'Authorization': `Bearer ${BEDDA_API_KEY}`,
    'Content-Type': 'application/json',
  },
  body: JSON.stringify({
    model: 'text-embedding-3-large', // or 'embed-english-v3', or others
    input: 'The text you want to embed',
  }),
});

The response format is consistent regardless of which model you use. Swap the model parameter to experiment with different embedding models without changing anything else in your pipeline.

Choosing the right embedding model

Different embedding models are better for different use cases:

Use case	Recommended model
General-purpose RAG	text-embedding-3-large
High-precision retrieval	text-embedding-3-large (3072 dims)
Cost-optimized at scale	text-embedding-3-small
Multilingual documents	Cohere embed-multilingual-v3
Open-source / self-hosted parity	bge-large-en-v1.5

With bedda.ai, switching between these is a one-line change in your code.

Try it

Access the embeddings API with your bedda.ai API key. The same key that powers your text and image AI access works for embeddings — no separate account needed.

---

This is part of the bedda.ai Feature Spotlight series. We publish a new feature every week.