OpenZync v1.0.0b3 — OpenRouter Embeddings

What's New

OpenRouter Embeddings

OpenRouter already handled chat completions — now it handles embeddings too. Use the same API key for both, configured independently via the org config:

• embedding_backend: "openrouter" — selectable from the dashboard embeddings page
• embedding_model: "nvidia/nemotron-3-embed-1b:free" — or any model OpenRouter proxies
• embedding_dim: 2048 — validated at runtime against the model output

The implementation uses raw HTTP (not the OpenAI SDK) because OpenRouter's response includes extra fields that the SDK parser rejects. Retry logic mirrors the chat backend: exponential backoff on 429/5xx, timeouts, and network errors.