AWS Bedrock + MemorySync

Overview

bedrock-memorysync ships the two patterns AWS teams actually deploy. MemoryConverseLoop wraps any bedrock-runtime client: before each converse call the user’s relevant memories are recalled under a hard 1.2s fail-open budget and injected as a system block, and both turns persist afterwards with deterministic idempotency seeds — memory can never delay or break the model call. The Bedrock Agents action group (Lambda + importable function schema + SAM template) gives any agent four curated memory functions — remember, recall_context, search_memories, forget_memory — with a multi-tenant identity ladder (sessionAttributes → parameter → sessionId), structured FAILURE response states instead of raw Lambda errors, and deliberately NO delete-everything function. No memory vendor ships a Bedrock Agents action-group integration — Mem0’s Bedrock story is an OSS sidecar requiring self-hosted Mem0 plus an OpenSearch domain you operate; Zep, Supermemory, and Letta have nothing. AWS’s own AgentCore Memory is the native alternative — at the cost of total AWS lock-in, black-box extraction strategies, and no cross-platform recall. MemorySync is one API key, and the memories flow to and from every other surface.

Setup status and requirements

Supported version

bedrock-memorysync 1.0.0 (PyPI)

Last setup review

2026-08-24

Permissions

A MemorySync API key with read and write scopes; AWS credentials for your own Bedrock access.

Limits

Python 3.10+; boto3 1.34+ for the Converse API. The action-group Lambda runs on python3.12.

Capabilities

• MemoryConverseLoop — recall as a system block, budgeted + fail-open
• The FIRST vendor Bedrock Agents action-group memory integration
• Multi-tenant identity ladder: sessionAttributes → parameter → sessionId
• Structured FAILURE responseState — never a raw Lambda error wall
• Deliberately NO delete-everything function
• Idempotent seeds — agent retries never duplicate
• SAM template + importable FUNCTION_SCHEMA for one-command deploys
• Cross-platform recall — unlike AgentCore’s AWS-only actorId

Quick Start

import boto3
from bedrock_memorysync import MemoryConverseLoop

loop = MemoryConverseLoop(
    bedrock_client=boto3.client("bedrock-runtime"),
    model_id="anthropic.claude-3-haiku-20240307-v1:0",
    user_id="customer-42",
    system_prompt="You are a helpful travel assistant.",
)
loop.chat("I always prefer window seats")
# a new session, days later:
loop.new_session("support-2")
loop.chat("which seat should I book for Oslo?")  # remembers

Use Cases

Explore More