Embeddings (semantic search)

4 embedding providers over HTTP, no SDK required:

Provider	Endpoint	Auth
OpenAI + compatible	`api.openai.com/v1/embeddings` or your own proxy	`Authorization: Bearer <key>`
Voyage AI	`api.voyageai.com/v1/embeddings`	same
Google Gemini	`generativelanguage.googleapis.com/...:batchEmbedContents`	key in the URL
Yandex Cloud	`llm.api.cloud.yandex.net/...`	`Authorization: Api-Key <key>` + folder ID

Configured entirely through the /embeddings web UI (provider, model, batch size).

Storage

A system table, __embeddings__, with columns (src_table, src_pk, model, vector, dims). The vector is stored as base64(big-endian float32); a 1536-dim OpenAI vector is roughly 10 KB per row.

Search

Cosine similarity, full scan. Tens of milliseconds for catalogs up to ~100k vectors.

API

# attach a table
curl -X POST $URL/api/embed/tables -H "X-Token: $TOK" -d '{
  "action": "upsert",
  "rule": {"table": "products", "text_expr": "CONCAT(name, ' — ', descr)", "enabled": true}
}'

# run indexing
curl -X POST $URL/api/embed/reindex -H "X-Token: $TOK" -d '{"table":"products"}'

# semantic search
curl -X POST $URL/api/embed/search -H "X-Token: $TOK" -d '{
  "text": "warm winter jacket", "table": "products", "k": 10
}'

Demo showcase mode and .fdb management

Features

← Previous

RBAC (multi-user)

File import