hypergraph-client

gRPC client for the Temporal Hypergraph Engine. The LangChain and LlamaIndex integrations are built on it.

pip install hypergraph-client
docker run -d -p 50051:50051 ghcr.io/divyansh151005/temporal-hypergraph-server

For development from a clone: pip install -e integrations/python-client.

from hypergraph_client import TemporalHypergraph

memory = TemporalHypergraph(api_key="your_key")  # session "default"; pass session_id= to change

from hypergraph_client import HypergraphClient, Hyperedge

with HypergraphClient("localhost:50051", api_key=None) as client:
    client.insert(Hyperedge("ACQUISITION", ["org:microsoft", "org:activision"], 1696118400,
                            summary="Microsoft completed its Activision acquisition"))
    for edge in client.traverse(["org:microsoft"], k_hops=3):
        print(edge.score, edge.summary)

• endpoint defaults to $HYPERGRAPH_ENDPOINT, then localhost:50051. It accepts host:port, tcp://host:port and unix:///path.
• api_key defaults to $HYPERGRAPH_API_KEY and is sent as x-api-key.
• AsyncHypergraphClient offers the same methods for asyncio.
• HypergraphKVOffloadTierManager stores and loads KV blocks over gRPC (submit_store, submit_load, close) with a vLLM-style interface. It is not yet loadable as a vLLM secondary tier; call it from your own code.
• SessionMemory stores chat messages around a session:<id> node; see the hypergraph_client.session module for the graph conventions.
• HashingEmbedder is a deterministic, dependency-free default embedder. Use a real embedding model for semantic retrieval quality.

Stubs in src/hypergraph_client/_proto/ are generated; regenerate them with scripts/gen_protos.sh python after editing proto/hypergraph_service.proto.

Tests (they start a local server; build it first with cargo build --release -p grpc-server):

pytest integrations/python-client