MixedBread

MixedBread produces domain-flexible sentence vectors that slot straight into SurrealDB’s built-in k-nearest-neighbour search (HNSW, brute-force or M-Tree). The result: a single database that stores your documents, graphs and semantic vectors—no extra services needed.

Install & run

pip install mixedbread surrealdb          # MixedBread client + async SurrealDB SDK
surreal start --log trace --auth root root   # local server (or connect to your cluster)

Generate embeddings

from mixedbread import MixedBreadModel

model = MixedBreadModel("mixedbread-variant")

sentences = [
    "SurrealDB merges SQL, graph and vector queries under one query language.",
    "MixedBread embeddings shine in e-commerce search scenarios.",
    "Vector similarity enables smarter product recommendations.",
]

embeddings = [model.embed(t) for t in sentences]    # returns 3 float arrays
DIM = len(embeddings[0])                            # auto-detect dimension

Ingest into SurrealDB

import asyncio
from surrealdb import Surreal

TABLE = "ProductDocs"

async def ingest():
    async with Surreal("ws://localhost:8000/rpc") as db:
        await db.signin({"username": "root", "password": "root"})
        await db.use("shop", "demo")             # namespace, database

        # ── schema + HNSW index (idempotent) ────────────────────────────────
        await db.query(`
            DEFINE TABLE IF NOT EXISTS {TABLE};
            DEFINE FIELD  text       ON {TABLE} TYPE string;
            DEFINE FIELD  embedding  ON {TABLE} TYPE array;
            DEFINE INDEX  hnsw_embed ON {TABLE}
                         FIELDS embedding HNSW DIMENSION {DIM};
        `)

        # ── insert rows ─────────────────────────────────────────────────────
        await db.create(TABLE, [
            {"id": f"doc:{i}", "text": txt, "embedding": vec}
            for i, (txt, vec) in enumerate(zip(sentences, embeddings))
        ])

asyncio.run(ingest())

Semantic search

async def search(query: str, k: int = 3):
    q_vec = model.embed(query)               # vectorise the query

    async with Surreal("ws://localhost:8000/rpc") as db:
        await db.signin({"username": "root", "password": "root"})
        await db.use("shop", "demo")

        result = await db.query(`
            LET $q = $vec;
            SELECT text,
                   vector::distance::cosine(embedding, $q) AS score
            FROM {TABLE}
            WHERE embedding <|{k}|> $q           -- top-k neighbours
            ORDER BY score;
        """, {"vec": q_vec})

        return result[0]["result"]

matches = asyncio.run(search("How do I unify SQL and vector search?"))
for m in matches:
    print(f"{m['score']:.4f}  {m['text']}")

Typical output:

0.1312  SurrealDB merges SQL, graph and vector queries under one query language.
0.4279  Vector similarity enables smarter product recommendations.
0.5127  MixedBread embeddings shine in e-commerce search scenarios.

Cheat-sheet

With MixedBread vectors inside SurrealDB, you get semantic search, recommendations and analytics from a single, lightweight engine—no separate vector store, no synchronisation headaches.