Getting Started

Agent Memory

Help an agent reuse durable constraints and confirmed outcomes across tasks while keeping current instructions, tool permissions, and live system state authoritative.

What an agent should remember

Store	Do not store as memory
User-approved working preferences	API keys, tokens, or tool credentials
Confirmed task outcomes	Unverified intermediate reasoning
Stable project constraints	Temporary tool output that is already stale
Reusable facts with source metadata	Instructions that can override the current system policy

Before a task

1Resolve the authenticated project and end-user scope.
2Query for constraints and prior outcomes relevant to the current task.
3Validate retrieved facts against live tools when freshness matters.
4Pass selected memory as untrusted context, separate from system instructions.

Retrieve task context

import os
from memorysync import MemorySyncClient

client = MemorySyncClient(
    api_key=os.environ["MEMORYSYNC_API_KEY"],
    base_url="https://api.memorysync.io",
    project_id=os.environ["MEMORYSYNC_PROJECT_ID"],
    end_user_id="user-123",
)

result = client.query("What constraints and prior decisions matter for this task?", k=5)
for memory in result.memories:
    print(memory.text)

After a task

Write only outcomes that completed successfully or were confirmed by the user.
Include source metadata that lets your application trace the outcome.
Update or remove stale constraints when the source of truth changes.
Keep tool authorization and execution logs in their appropriate systems.

Agent safety checks

Retrieved memory cannot grant tool access.
Current system and user instructions outrank retrieved context.
Sensitive actions require live authorization and confirmation.
A failed or partial task is not stored as a successful outcome.

← Previous

Cost Optimization

RAG & Knowledge Assistants