openmemory-py 1.2.3

OpenMemory Python SDK provides a local-first long-term memory engine for AI agents and LLM applications. Features include semantic search, multi-sector memory, temporal fact storage, automatic decay, and explainable recall paths. Works fully offline or with the OpenMemory backend.

OpenMemory Python SDK

VS Code Extension • Report Bug • Request Feature • Discord

Local-first long-term memory engine for AI apps and agents. Self-hosted. Explainable. Scalable.

---

Quick Start

pip install openmemory-py

from openmemory import OpenMemory

mem = OpenMemory(
    path='./data/memory.sqlite',
    tier='fast',
    embeddings={
        'provider': 'synthetic'  # or 'openai', 'gemini', 'ollama'
    }
)

mem.add("I'm building a Django app with OpenMemory")
results = mem.query("What am I building?")
print(results)

That's it. You're now running a fully local cognitive memory engine 🎉

---

Features

✅ Local-first - Runs entirely on your machine, zero external dependencies
✅ Multi-sector memory - Episodic, Semantic, Procedural, Emotional, Reflective
✅ Temporal knowledge graph - Time-aware facts with validity periods
✅ Memory decay - Adaptive forgetting with sector-specific rates
✅ Waypoint graph - Associative recall paths for better retrieval
✅ Explainable traces - See exactly why memories were recalled
✅ Zero config - Works out of the box with sensible defaults

---

Configuration

Required Configuration

All three parameters are required for local mode:

mem = OpenMemory(
    path='./data/memory.sqlite',      # Where to store the database
    tier='fast',                       # Performance tier
    embeddings={
        'provider': 'synthetic'         # Embedding provider
    }
)

Embedding Providers

Synthetic (Testing/Development)

embeddings={'provider': 'synthetic'}

OpenAI (Recommended for Production)

import os

embeddings={
    'provider': 'openai',
    'apiKey': os.getenv('OPENAI_API_KEY'),
    'model': 'text-embedding-3-small'  # optional
}

Gemini

embeddings={
    'provider': 'gemini',
    'apiKey': os.getenv('GEMINI_API_KEY')
}

Ollama (Fully Local)

embeddings={
    'provider': 'ollama',
    'model': 'llama3',
    'ollama': {
        'url': 'http://localhost:11434'  # optional
    }
}

AWS Bedrock

embeddings={
    'provider': 'aws',
    'aws': {
        'accessKeyId': os.getenv('AWS_ACCESS_KEY_ID'),
        'secretAccessKey': os.getenv('AWS_SECRET_ACCESS_KEY'),
        'region': 'us-east-1'
    }
}

Performance Tiers

• fast - Optimized for speed, lower precision
• smart - Balanced performance and accuracy
• deep - Maximum accuracy, slower
• hybrid - Adaptive based on query complexity

Advanced Configuration

mem = OpenMemory(
    path='./data/memory.sqlite',
    tier='smart',
    embeddings={
        'provider': 'openai',
        'apiKey': os.getenv('OPENAI_API_KEY')
    },
    decay={
        'intervalMinutes': 60,
        'reinforceOnQuery': True,
        'coldThreshold': 0.1
    },
    compression={
        'enabled': True,
        'algorithm': 'semantic',
        'minLength': 100
    },
    reflection={
        'enabled': True,
        'intervalMinutes': 10,
        'minMemories': 5
    }
)

---

API Reference

add(content, **options)

Store a new memory.

result = mem.add(
    "User prefers dark mode",
    tags=["preference", "ui"],
    metadata={"category": "settings"},
    decayLambda=0.01  # Custom decay rate
)

query(query, **options)

Search for relevant memories.

results = mem.query("user preferences", limit=10, minScore=0.7)

getAll(**options)

Retrieve all memories.

all_memories = mem.getAll(limit=100, offset=0)

getBySector(sector, **options)

Get memories from a specific cognitive sector.

episodic = mem.getBySector('episodic', limit=20)
semantic = mem.getBySector('semantic')

Available sectors: episodic, semantic, procedural, emotional, reflective

delete(id)

Remove a memory by ID.

mem.delete(memory_id)

close()

Close the database connection (important for cleanup).

mem.close()

---

Cognitive Sectors

OpenMemory automatically classifies content into 5 cognitive sectors:

Sector	Description	Examples	Decay Rate
Episodic	Time-bound events & experiences	"Yesterday I attended a conference"	Medium
Semantic	Timeless facts & knowledge	"Paris is the capital of France"	Very Low
Procedural	Skills, procedures, how-tos	"To deploy: build, test, push"	Low
Emotional	Feelings, sentiment, mood	"I'm excited about this project!"	High
Reflective	Meta-cognition, insights	"I learn best through practice"	Very Low

---

Examples

Check out the examples/py-sdk/ directory for comprehensive examples:

• basic_usage.py - CRUD operations
• advanced_features.py - Decay, compression, reflection
• brain_sectors.py - Multi-sector demonstration
• performance_benchmark.py - Performance testing

---

Remote Mode

For production deployments with a centralized OpenMemory server:

mem = OpenMemory(
    mode='remote',
    url='https://your-backend.com',
    apiKey='your-api-key'
)

---

Performance

• 115ms average recall @ 100k memories
• 338 QPS throughput with 8 workers
• 95% recall accuracy @ k=5
• 7.9ms/item scoring at 10k+ scale

---

Type Hints

Full type hint support included:

from typing import List, Dict, Any
from openmemory import OpenMemory

mem: OpenMemory = OpenMemory(
    path='./data/memory.sqlite',
    tier='fast',
    embeddings={'provider': 'synthetic'}
)

results: List[Dict[str, Any]] = mem.query("test")

---

License

Apache 2.0

---

Links

• Main Repository
• Documentation
• Examples
• VS Code Extension