strands-memmachine 0.1.0

MemMachine memory tool for Strands Agents

strands-memmachine

MemMachine memory tool for Strands Agents

Strands Docs ◆ MemMachine ◆ API Docs ◆ Community Packages

A Strands Agents extension that integrates MemMachine as a persistent memory backend, enabling AI agents to store, search, list, and delete episodic and semantic memories across sessions. It also supports full project management — create, retrieve, list, and delete isolated memory namespaces.

What is MemMachine?

MemMachine is an AI memory platform by MemVerge that provides persistent, structured memory for AI agents. It enables agents to remember information across sessions, learn from past interactions, and deliver personalized, context-aware responses.

MemMachine supports two memory types:

• Episodic memory — conversational, event-based memories tied to interactions
• Semantic memory — factual, structured knowledge extracted from conversations

Memory is organized into Projects — isolated namespaces that keep memories separated by use case, user, or application.

Installation

pip install strands-memmachine

Configuration

Environment Variable	Required	Description
MEMMACHINE_API_KEY	Yes	API key from console.memmachine.ai

export MEMMACHINE_API_KEY=your_api_key

Usage

from strands import Agent
from strands_memmachine import memmachine_memory

agent = Agent(tools=[memmachine_memory])

# Store a memory
agent.tool.memmachine_memory(
    action="store",
    content="User prefers aisle seats on flights",
    metadata={"user_id": "alice", "category": "travel"},
)

# Search memories
agent.tool.memmachine_memory(
    action="search",
    query="What are the flight preferences?",
    top_k=5,
)

# List memories with filter
agent.tool.memmachine_memory(
    action="list",
    filter="metadata.user_id=alice AND metadata.category=travel",
    page_size=20,
)

# Delete a single episodic memory
agent.tool.memmachine_memory(
    action="delete",
    memory_type="episodic",
    memory_id="mem-123",
)

# Bulk delete semantic memories
agent.tool.memmachine_memory(
    action="delete",
    memory_type="semantic",
    memory_ids=["sem-1", "sem-2", "sem-3"],
)

# Create a project
agent.tool.memmachine_memory(
    action="create_project",
    project_id="my-project",
    description="Project for storing travel preferences",
)

# Get a project
agent.tool.memmachine_memory(
    action="get_project",
    project_id="my-project",
)

# List all projects
agent.tool.memmachine_memory(
    action="list_projects",
)

# Get episode count for a project
agent.tool.memmachine_memory(
    action="get_episode_count",
    project_id="my-project",
)

# Delete a project
agent.tool.memmachine_memory(
    action="delete_project",
    project_id="my-project",
)

Actions

Memory Actions

Action	Required Parameters	Optional Parameters
store	content	types, producer, produced_for, metadata
search	query	top_k, types, filter
list	—	memory_type, filter, page_size, page_num
delete	memory_type, memory_id or memory_ids	—

Project Actions

Action	Required Parameters	Optional Parameters
create_project	project_id	description
delete_project	project_id	—
get_project	project_id	—
list_projects	—	—
get_episode_count	project_id	—

Memory Types

• Episodic — conversational, event-based memories
• Semantic — factual, structured knowledge

Use the types parameter to target one or both:

agent.tool.memmachine_memory(
    action="store",
    content="User is a software engineer",
    types=["semantic"],
)

Metadata Filter Syntax

Used with search and list actions:

# Single condition
metadata.user_id=alice

# Multiple conditions
metadata.user_id=alice AND metadata.category=travel

Parameters

Memory Parameters

Parameter	Type	Description
action	str	Operation to perform
content	str	Memory text. Required for store
query	str	Search query. Required for search
memory_id	str	Single UID to delete. Required for delete when memory_ids not provided
memory_ids	list[str]	List of UIDs for bulk delete. Required for delete when memory_id not provided
memory_type	str	episodic or semantic. Required for delete, optional for list
types	list[str]	Memory types to write to or search. Defaults to both when omitted
top_k	int	Max search results. Default: 10
page_size	int	Results per page for list. Default: 100
page_num	int	Zero-based page index for list. Default: 0
filter	str	Metadata filter expression for search and list
producer	str	Producer identity label for store. Default: user
produced_for	str	Intended recipient for store
metadata	dict	Key-value pairs attached to the stored memory. All values coerced to strings

Project Parameters

Parameter	Type	Description
project_id	str	Unique project identifier. Allowed characters: letters, numbers, underscores, hyphens, colons, and Unicode
description	str	Optional human-readable description of the project

Resources

• MemMachine Platform
• MemMachine API Docs
• MemVerge
• Strands Agents Documentation
• Strands Community Packages
• Get Featured in Strands Docs