Agno + MemorySync

Overview

Install agno-memorysync and wire MemoryManager(db=MemorySyncDb()) while sessions stay in your local db — the backend implements every memory method for real and makes every other BaseDb surface raise with the fix in the message, so silent session loss is impossible. Memories extracted by Agno’s native pipeline (update_memory_on_run, agentic memory, image-turn understanding) are kept in MemorySync framework state, one value per memory id, and injected into context on every run; each memory’s text is also sent to fact extraction, so its facts are recallable from every MemorySync surface. get_user_memories(search_content=...) ranks this user’s memories by similarity server-side — Agno’s built-in retrieval is last_n, first_n, or an extra LLM round-trip, and the Mem0 alternative is a toolkit the model must remember to call, with a sync client that blocks async agents and a cookbook that instructs you to comment out the add call after the first run because nothing is idempotent. Here an unchanged memory is recognised and never stored or extracted twice (retries and re-runs converge), an update replaces the facts of the previous version, a delete removes the memory and its facts, reads run under a hard 1.2s budget and fail open, writes fail open with loud logs, failed deletes raise, and the nullary clear_memories() always refuses — per-user wipes are explicit. An AsyncMemorySyncDb twin rides agno’s native AsyncBaseDb path for Agent.arun.

Setup status and requirements

Supported version

agno-memorysync 1.1.0 (PyPI)

Last setup review

2026-08-31

Permissions

A MemorySync API key with read and write scopes.

Limits

Requires agno 2.8+ (Python 3.10+).

Capabilities

• Memory-only BaseDb — 33 non-memory surfaces raise with guidance, never silent loss
• Server-side similarity search via search_content
• AsyncMemorySyncDb twin — native AsyncBaseDb, no blocked event loop
• One value per memory id — an unchanged memory is never stored or extracted twice
• Each memory’s text becomes facts; updates replace them, deletes remove them
• clear_memories() always refuses; forget_user_memories(user_id) is the explicit wipe
• Reads fail open under a 1.2s budget; writes fail open with loud logs; deletes raise
• agent_id / team_id / topics stored and filterable
• Image-turn memories through agno’s NATIVE pipeline — no bypass

Quick Start

from agno.agent import Agent
from agno.db.sqlite import SqliteDb
from agno.memory import MemoryManager
from agno_memorysync import MemorySyncDb

agent = Agent(
    db=SqliteDb(db_file="agent.db"),                  # sessions: local
    memory_manager=MemoryManager(db=MemorySyncDb()),  # memories: MemorySync
    update_memory_on_run=True,
    user_id="customer-42",
)
agent.run("I prefer teal dashboards and window seats")
agent.run("Which color should the new chart use?")   # remembers

Use Cases

Explore More