onemillionbrain 0.1.10

A portable, semantic graph memory layer for any AI agent.

1MBrain Python SDK

Python client for the 1MBrain REST API — a portable, semantic graph memory layer for AI agents.

Installation

# Sync client (zero extra dependencies — uses stdlib urllib)
pip install onemillionbrain

# Async client (requires httpx)
pip install onemillionbrain[async]

Quick Start

Sync

from onemillionbrain import OneMBrainClient

client = OneMBrainClient(
    api_url="http://localhost:3001",
    api_key="your-api-key",
    agent_id="my-agent",
)

# Store a memory
memory = client.remember("User prefers Bahasa Indonesia as primary language", type="semantic")
print(memory.id)

# Search memories
results = client.recall("language preference", limit=5)
for r in results:
    print(f"[{r.score:.3f}] {r.memory.content}")

# Create an explicit association
client.associate(results[0].memory.id, results[1].memory.id, strength=0.8)

# Forget a memory
client.forget(memory.id)

Async

import asyncio
from onemillionbrain import AsyncOneMBrainClient

async def main():
    async with AsyncOneMBrainClient(
        api_url="http://localhost:3001",
        api_key="your-api-key",
        agent_id="my-agent",
    ) as client:
        memory = await client.remember("User asked about pricing on 2026-06-10", type="episodic")
        results = await client.recall("pricing questions")
        await client.forget(memory.id)

asyncio.run(main())

Agent Integration

To ensure your LLM agent knows exactly how and when to use 1MBrain, the SDK exports a pre-written AGENT_SYSTEM_PROMPT. Inject this into your agent's system instructions.

from onemillionbrain import AGENT_SYSTEM_PROMPT

system_instruction = f"""
You are a helpful AI assistant.
{AGENT_SYSTEM_PROMPT}
"""

# Pass system_instruction to LangChain, OpenAI, Anthropic, etc.

LangChain Integration

from langchain.tools import tool
from onemillionbrain import OneMBrainClient

brain = OneMBrainClient(
    api_url="http://localhost:3001",
    api_key="your-api-key",
    agent_id="langchain-agent",
)

@tool
def remember_tool(content: str) -> str:
    """Store something in long-term memory."""
    memory = brain.remember(content, type="episodic")
    return f"Stored memory: {memory.id}"

@tool
def recall_tool(query: str) -> str:
    """Search long-term memory."""
    results = brain.recall(query, limit=5)
    if not results:
        return "No memories found."
    return "\n".join(f"- {r.memory.content}" for r in results)

API Reference

OneMBrainClient(api_url, api_key, agent_id=None)

Parameter	Type	Description
api_url	str	Base URL of the 1MBrain API (e.g. http://localhost:3001)
api_key	str	Your API key (passed as X-API-Key header)
agent_id	str	Default agent namespace (can be overridden per call)

Methods

Method	Signature	Returns
remember	(content, *, type, importance, tags, metadata, agent_id)	Memory
recall	(query, *, limit, type, tags, max_hops, activation_threshold, blend_weight, agent_id, cross_agent)	list[RecallResult]
forget	(memory_id, *, agent_id)	bool
associate	(source_id, target_id, *, strength, origin, agent_id)	AssociateResult

Development

pip install -e ".[dev]"
pytest