langgraph-synapse

An integration package connecting Synapse memory and LangGraph.

Privacy-first checkpointing and cross-thread memory — all data stays local.

Installation

pip install langgraph-synapse

Components

SynapseCheckpointer

Persist LangGraph state across invocations:

from synapse import Synapse
from langgraph_synapse import SynapseCheckpointer

syn = Synapse("./agent_state")
checkpointer = SynapseCheckpointer(synapse=syn)

# Use with LangGraph
graph = builder.compile(checkpointer=checkpointer)
graph.invoke(
    {"messages": [{"role": "user", "content": "hello"}]},
    config={"configurable": {"thread_id": "thread-1"}},
)

SynapseStore

Cross-thread memory store with semantic search:

from synapse import Synapse
from langgraph_synapse import SynapseStore

syn = Synapse("./shared_memory")
store = SynapseStore(synapse=syn)

# Store shared context
store.put(("user", "preferences"), "diet", {"value": "vegetarian"})

# Retrieve across any thread
item = store.get(("user", "preferences"), "diet")

# Semantic search
results = store.search(("user",), query="food preferences")

Why Synapse?

• 🔒 Privacy-first: All state stays on your machine
• 🧠 Semantic recall: Not just key-value — find relevant state by meaning
• ⚡ Zero dependencies: Synapse is pure Python
• 📦 Portable: .synapse files can be shared and federated

License

MIT