llama-index-vector-stores-synapcores

LlamaIndex vector store integration for SynapCores — the AI-native SQL engine that combines vectors, graph, SQL, and AI inference in one self-hosted binary.

pip install llama-index llama-index-vector-stores-synapcores

Quickstart

from llama_index.core import VectorStoreIndex, StorageContext
from llama_index.core.readers import SimpleDirectoryReader
from llama_index.vector_stores.synapcores import SynapCoresVectorStore

# 1. Start the engine: docker run -p 8080:8080 -e AIDB_ACCEPT_LICENSE=1 synapcores/community:latest
# 2. Point the integration at it
vector_store = SynapCoresVectorStore(
    uri="http://localhost:8080",
    table_name="my_docs",
    embedding_dim=1536,           # match your embedding model
)

# 3. Standard LlamaIndex flow — no per-vendor weirdness
docs = SimpleDirectoryReader("./data").load_data()
index = VectorStoreIndex.from_documents(
    docs,
    storage_context=StorageContext.from_defaults(vector_store=vector_store),
)
response = index.as_query_engine().query("What does this document say?")
print(response)

Configuration

Metadata filters

LlamaIndex's full filter grammar maps to SynapCores' JSON arrow operator:

from llama_index.core.vector_stores import MetadataFilters, MetadataFilter, FilterOperator

filters = MetadataFilters(filters=[
    MetadataFilter(key="tenant_id", value="acme", operator=FilterOperator.EQ),
    MetadataFilter(key="doc_type", value=["policy", "contract"], operator=FilterOperator.IN),
])

retriever = index.as_retriever(filters=filters)

Supported operators: EQ, NE, GT, GTE, LT, LTE, IN, NIN, TEXT_MATCH, TEXT_MATCH_INSENSITIVE, CONTAINS, IS_EMPTY. Conditions: AND, OR, NOT. Filter groups may be nested. ANY / ALL (array-membership) are tracked for v0.2.0.

Existing store

If you've already ingested:

vector_store = SynapCoresVectorStore(uri="http://localhost:8080", table_name="my_docs", embedding_dim=1536)
index = VectorStoreIndex.from_vector_store(vector_store)

Do I need to install any models in SynapCores?

No. The integration works out-of-the-box against a stock synapcores/community:latest container with zero model downloads.

Embeddings are computed by LlamaIndex — OpenAIEmbedding, HuggingFaceEmbedding, CohereEmbedding, or whatever you've configured in your LlamaIndex Settings.embed_model. The integration writes the resulting List[float] to a VECTOR(N) column. SynapCores just stores it. There is no engine-side embedding model to download, no GGUF, no Ollama, no API key, nothing.

If you separately want to use the engine's in-DB AI surface (SELECT EMBED(...), SELECT GENERATE(...), SELECT AGENT_RUN(...)) — those need a model warmed via synapcores pull <model:tag> or an [query.ai_service] block in gateway.toml. But for this integration alone, you can stop reading here: it works against a fresh docker run with no further setup.

Engine requirements

• synapcores/community:v1.7.0.2-ce or newer

Async

async_add, aquery, adelete, adelete_nodes, aclear are available. v0.1.0 wraps the sync client via asyncio.to_thread — non-blocking but uses a thread per call. Native httpx path is on the v0.2.0 roadmap.

License

MIT