SharedMemory Python SDK

The persistent memory layer for AI agents. Add, search, and manage long-term memories with knowledge graph, entity scoping, session lifecycle, and structured extraction.

Installation

pip install sharedmemory-ai

Quick Start

from sharedmemory import SharedMemory

memory = SharedMemory(
    api_key="sm_live_...",
    volume_id="your-volume-uuid",
)

# Store a memory
result = memory.remember("User prefers dark mode and compact layout")
print(result["status"])  # "approved"

# Query memories
results = memory.query("what are the user's UI preferences?")
for source in results["sources"]:
    print(source["content"], source["score"])

# Chat (RAG + LLM answer) — the default way to interact
result = memory.chat("What does the user prefer for their UI?")
print(result["answer"])
print(result["sources"])

# Batch write
memory.remember_many([
    {"content": "User's name is Alice"},
    {"content": "Alice works at Acme Corp"},
    {"content": "Alice prefers Python over JavaScript"},
])

Entity Scoping

Scope memories to specific users, agents, or sessions:

memory = SharedMemory(
    api_key="sm_live_...",
    volume_id="vol-uuid",
    user_id="user-123",
    agent_id="chatbot-1",
)

# All operations are automatically scoped
memory.remember("User asked about pricing", session_id="sess-abc")
results = memory.query("pricing", session_id="sess-abc")

Sessions

Track conversation sessions with automatic summarization:

# Start a session
memory.start_session("conv-001", user_id="user-123")

# Add memories during the session
memory.remember("User asked about pricing", session_id="conv-001")
memory.remember("User is interested in enterprise plan", session_id="conv-001")

# End session — automatically summarizes into long-term memory
summary = memory.end_session("conv-001", auto_summarize=True)

Advanced Search

# With reranking
results = memory.query(
    "project deadlines",
    rerank=True,
    rerank_method="llm",
    include_context=True,
)

# With metadata filters
results = memory.query("preferences", filters={
    "AND": [
        {"field": "memory_class", "op": "eq", "value": "preference"},
        {"field": "score", "op": "gte", "value": 0.5},
    ]
})

Knowledge Graph

# Get entity details
entity = memory.get_entity("Alice")
print(entity["summary"])
print(entity["facts"])

# Get full graph
graph = memory.get_graph()

Feedback

memory.feedback("memory-uuid", "POSITIVE", reason="Highly relevant")
memory.feedback("memory-uuid", "NEGATIVE", reason="Outdated information")

Context Assembly

Get an optimized context block for LLM prompting:

context = memory.get_context(template_id="conversational")
print(context["blocks"])  # Ready for system prompt injection

Structured Extraction

data = memory.extract(
    text="Alice is 28, works at Acme Corp as a Senior Engineer since 2022",
    schema_id="contact-info",
)
print(data)  # {"name": "Alice", "age": 28, "company": "Acme Corp", ...}

Async Client

from sharedmemory.async_client import AsyncSharedMemory

async with AsyncSharedMemory(api_key="sm_live_...") as memory:
    await memory.remember("async memory")
    results = await memory.query("query")

API Reference

License

MIT