Proxy API
Embeddings
The OpenAI-compatible embeddings surface on the MarkDB proxy.
POST /v1/embeddings on the proxy is
OpenAI-compatible and, like the other surfaces, is routed by model
name to the provider that owns it.
Request
curl https://proxy.markdb.cloud/v1/embeddings \
-H "Authorization: Bearer mk_live_xxx" \
-H "Content-Type: application/json" \
-d '{
"model": "gemini-embedding-001",
"input": "MarkDB is the memory database for AI agents.",
"dimensions": 3072
}'inputaccepts a string or an array of strings.dimensionsis optional -- embedding models expose a Matryoshka grid of legal sizes; omit it to use the model default.
Response
The standard OpenAI embeddings shape:
{
"object": "list",
"data": [
{ "object": "embedding", "index": 0, "embedding": [0.0123, -0.0456, "…"] }
],
"model": "gemini-embedding-001",
"usage": { "prompt_tokens": 9, "total_tokens": 9 }
}Same endpoint powers search
MarkDB's own hybrid search indexer embeds your summaries through this exact endpoint, so query-time and index-time vectors share a model and dimension. Keep the embedder config consistent and search stays comparable.