ai-infra / API Reference

MemoryStore

from ai_infra.llm.memory import MemoryStore

ai_infra.llm.memory

Long-term memory store with semantic search. Provides a simple interface for storing and retrieving memories with optional semantic search capabilities. Supports multiple storage backends: - In-memory (dev/testing) - SQLite (single-instance) - PostgreSQL (production) Memories are organized by namespace (tuple of strings) and key. Namespaces enable multi-user and multi-tenant storage.

python

from ai_infra.memory import MemoryStore

    # Simple in-memory store
    store = MemoryStore()

    # With semantic search (requires embedding provider)
    store = MemoryStore(
        embedding_provider="openai",
        embedding_model="text-embedding-3-small",
    )

    # Store user preferences
    store.put(
        namespace=("user_123", "preferences"),
        key="language",
        value={"preference": "User prefers Python"},
    )

    # Search memories
    results = store.search(
        ("user_123", "preferences"),
        query="what programming language",
        limit=5,
    )

Constructor

MemoryStore(backend: str = 'memory', embedding_provider: str | None = None, embedding_model: str | None = None, connection_string: str | None = None, path: str | None = None)

Parameter	Type	Default	Description
`backend`	`str`	'memory'	Storage backend ("memory", "sqlite", "postgres")
`embedding_provider`	`str\|None`	None	Provider for semantic search (e.g., "openai", "huggingface")
`embedding_model`	`str\|None`	None	Model for embeddings (e.g., "text-embedding-3-small")
`connection_string`	`str\|None`	None	PostgreSQL connection string (for postgres backend)
`path`	`str\|None`	None	File path (for sqlite backend)

Methods

from ai_infra.memory import MemoryStore # Simple in-memory store store = MemoryStore() # With semantic search (requires embedding provider) store = MemoryStore( embedding_provider="openai", embedding_model="text-embedding-3-small", ) # Store user preferences store.put( namespace=("user_123", "preferences"), key="language", value={"preference": "User prefers Python"}, ) # Search memories results = store.search( ("user_123", "preferences"), query="what programming language", limit=5, )

Parameter

Type

Default

Description

backend

str

'memory'

Storage backend ("memory", "sqlite", "postgres")

embedding_provider

str|None

None

Provider for semantic search (e.g., "openai", "huggingface")

embedding_model

str|None

None

Model for embeddings (e.g., "text-embedding-3-small")

connection_string

str|None

None

PostgreSQL connection string (for postgres backend)

path

str|None

None

File path (for sqlite backend)