Configuration¶

GraphMem is configured using the MemoryConfig class.

Basic Configuration¶

from graphmem import MemoryConfig

config = MemoryConfig(
    # Required: LLM provider
    llm_provider="openai",
    llm_api_key="sk-...",
    llm_model="gpt-4o-mini",

    # Required: Embedding provider
    embedding_provider="openai",
    embedding_api_key="sk-...",
    embedding_model="text-embedding-3-small",
)

LLM Configuration¶

OpenAI¶

config = MemoryConfig(
    llm_provider="openai",
    llm_api_key="sk-...",
    llm_model="gpt-4o-mini",  # or gpt-4o, gpt-4-turbo
)

Azure OpenAI¶

config = MemoryConfig(
    llm_provider="azure_openai",
    llm_api_key="your-azure-key",
    llm_api_base="https://your-resource.openai.azure.com/",  # Azure endpoint
    azure_deployment="gpt-4",
    llm_model="gpt-4",
    azure_api_version="2024-02-15-preview",  # Optional
)

OpenAI-Compatible (OpenRouter, Together, Groq, etc.)¶

config = MemoryConfig(
    llm_provider="openai_compatible",
    llm_api_key="your-key",
    llm_api_base="https://openrouter.ai/api/v1",  # Custom base URL
    llm_model="google/gemini-2.0-flash-001",

    embedding_provider="openai_compatible",
    embedding_api_key="your-key",
    embedding_api_base="https://openrouter.ai/api/v1",  # Custom base URL
    embedding_model="openai/text-embedding-3-small",
)

Local LLMs (Ollama, vLLM, LM Studio)¶

config = MemoryConfig(
    llm_provider="openai_compatible",
    llm_api_key="not-needed",  # Local models often don't need keys
    llm_api_base="http://localhost:11434/v1",  # Ollama
    llm_model="llama3.2",

    embedding_provider="openai_compatible",
    embedding_api_key="not-needed",
    embedding_api_base="http://localhost:11434/v1",
    embedding_model="nomic-embed-text",
)

Anthropic¶

config = MemoryConfig(
    llm_provider="anthropic",
    llm_api_key="sk-ant-...",
    llm_model="claude-3-5-sonnet-20241022",
)

Embedding Configuration¶

OpenAI Embeddings¶

config = MemoryConfig(
    embedding_provider="openai",
    embedding_api_key="sk-...",
    embedding_model="text-embedding-3-small",  # or text-embedding-3-large
)

Azure OpenAI Embeddings¶

config = MemoryConfig(
    embedding_provider="azure_openai",
    embedding_api_key="your-azure-key",
    embedding_api_base="https://your-resource.openai.azure.com/",  # Azure endpoint
    azure_embedding_deployment="text-embedding-ada-002",
    embedding_model="text-embedding-ada-002",
    azure_api_version="2024-02-15-preview",  # Optional
)

Storage Configuration¶

In-Memory (Default)¶

config = MemoryConfig(
    # No storage config needed - uses in-memory by default
)

Turso (SQLite Persistence)¶

Important

turso_db_path is required to enable Turso. Just providing turso_url without turso_db_path will NOT use Turso!

# Local-only (no cloud sync)
config = MemoryConfig(
    turso_db_path="my_memory.db",  # ✅ REQUIRED - local SQLite file
)

# With cloud sync (local + cloud backup)
config = MemoryConfig(
    turso_db_path="my_memory.db",  # ✅ REQUIRED - local SQLite file
    turso_url="libsql://your-db.turso.io",  # Optional - cloud sync URL
    turso_auth_token="your-token",  # Optional - cloud auth token
)

Turso uses an offline-first architecture: all data is stored locally, with optional cloud sync for backup and multi-device access. See Storage Backends for details.

Neo4j¶

config = MemoryConfig(
    neo4j_uri="neo4j+s://xxx.databases.neo4j.io",
    neo4j_username="neo4j",
    neo4j_password="your-password",
)

Redis Cache¶

config = MemoryConfig(
    redis_url="redis://default:password@host:port",
)

Evolution Configuration¶

config = MemoryConfig(
    # Enable memory evolution
    evolution_enabled=True,

    # Auto-evolve after each ingestion
    auto_evolve=True,

    # Memory decay settings
    decay_enabled=True,
    decay_half_life_days=30,  # How fast memories fade

    # Consolidation threshold (0.0 - 1.0)
    consolidation_threshold=0.85,  # Higher = less aggressive merging
)

Multi-Tenant Configuration¶

from graphmem import GraphMem

# User-specific memory
memory = GraphMem(
    config,
    user_id="alice",      # Isolate by user
    memory_id="chat_1",   # Isolate by session/context
)

All Configuration Options¶

Parameter	Type	Default	Description
`llm_provider`	str	Required	LLM provider name
`llm_api_key`	str	Required	API key for LLM
`llm_model`	str	Required	Model name
`llm_api_base`	str	None	Custom LLM API base URL (for OpenRouter, Together, local models)
`embedding_provider`	str	Required	Embedding provider
`embedding_api_key`	str	Required	API key for embeddings
`embedding_model`	str	Required	Embedding model name
`embedding_api_base`	str	None	Custom embedding API base URL (for OpenRouter, Together, local models)
`azure_endpoint`	str	None	Azure OpenAI endpoint
`azure_deployment`	str	None	Azure LLM deployment
`azure_embedding_deployment`	str	None	Azure embedding deployment
`azure_api_version`	str	"2024-02-15-preview"	Azure API version
`turso_db_path`	str	None	Required for Turso - Local SQLite file path
`turso_url`	str	None	Optional - Turso cloud URL for sync (requires `turso_db_path`)
`turso_auth_token`	str	None	Optional - Turso auth token (requires `turso_url`)
`neo4j_uri`	str	None	Neo4j connection URI
`neo4j_username`	str	None	Neo4j username
`neo4j_password`	str	None	Neo4j password
`redis_url`	str	None	Redis connection URL
`evolution_enabled`	bool	True	Enable evolution
`auto_evolve`	bool	False	Auto-evolve on ingest
`decay_enabled`	bool	True	Enable memory decay
`decay_half_life_days`	int	30	Decay half-life
`consolidation_threshold`	float	0.85	Merge threshold

Environment Variables¶

You can also use environment variables:

export OPENAI_API_KEY="sk-..."
export NEO4J_URI="neo4j+s://..."
export NEO4J_PASSWORD="..."
export REDIS_URL="redis://..."

import os

config = MemoryConfig(
    llm_provider="openai",
    llm_api_key=os.getenv("OPENAI_API_KEY"),
    llm_model="gpt-4o-mini",
    embedding_provider="openai",
    embedding_api_key=os.getenv("OPENAI_API_KEY"),
    embedding_model="text-embedding-3-small",
    neo4j_uri=os.getenv("NEO4J_URI"),
    neo4j_password=os.getenv("NEO4J_PASSWORD"),
)