rem-memory 0.1.0

Recursive Episodic Memory for AI Agents — persistent memory infrastructure

REM Python SDK

rem-memory is the official Python SDK for REM (Recursive Episodic Memory) — an open-source memory layer for AI agents.

---

Installation

pip install rem-memory

Quick start

import asyncio
from rem_memory import REMClient

async def main():
    async with REMClient(api_key="rem_sk_...") as client:
        # Write a memory episode
        result = await client.write(
            content="User prefers TypeScript with strict mode for all new projects.",
            agent_id="agent_cursor",
            user_id="user_alice",
        )
        print(result.episode_id)

        # Retrieve relevant memories
        memories = await client.retrieve(
            query="Does the user prefer TypeScript or JavaScript?",
            agent_id="agent_cursor",
            top_k=5,
        )
        for m in memories.episodes:
            print(m.intent, m.importance_score)

asyncio.run(main())

Configuration

Parameter	Default	Description
api_key	—	Your REM API key (X-API-Key header)
base_url	http://localhost:8080	Base URL of the REM Go API
timeout	30	Request timeout in seconds

Environment variables are also supported:

export REM_API_KEY=rem_sk_...
export REM_BASE_URL=http://localhost:8080

Integrations

LangChain

from rem_memory.integrations.langchain import REMMemory
from langchain.chains import ConversationChain
from langchain_openai import ChatOpenAI

memory = REMMemory(
    api_key="rem_sk_...",
    agent_id="agent_cursor",
    user_id="user_alice",
)

chain = ConversationChain(llm=ChatOpenAI(), memory=memory)
response = chain.predict(input="What are my coding preferences?")

AutoGen

from rem_memory.integrations.autogen import REMMemoryStore
from autogen import ConversableAgent

memory_store = REMMemoryStore(
    api_key="rem_sk_...",
    agent_id="agent_autogen",
)

agent = ConversableAgent(
    name="assistant",
    system_message="You are a helpful coding assistant.",
)

# Inject past memories into system prompt
context = await memory_store.get_context(query="user coding preferences")

API Reference

REMClient

write(content, agent_id, user_id, session_id?, metadata?)

Write a raw interaction to episodic memory. Returns WriteResult.

retrieve(query, agent_id, top_k?, include_semantic?)

Retrieve semantically relevant episodes and facts. Returns RetrieveResult.

get_agent(agent_id)

Fetch agent metadata. Returns Agent.

list_agents()

List all agents. Returns List[Agent].

list_episodes(agent_id, limit?, offset?)

Page through raw episodes for an agent. Returns List[Episode].

list_semantic_memories(agent_id)

Fetch all consolidated semantic memories. Returns List[SemanticMemory].

Development

cd sdk/python
pip install -e ".[dev]"
pytest

License

MIT © REM Contributors