ag2-memorysync

MemorySync for AG2 (the classic AutoGen ConversableAgent framework, pip install autogen): an automatic memory loop — recall injected before every reply, both sides persisted, zero extra code per turn.

pip install ag2-memorysync

Quick start

from autogen import ConversableAgent
from ag2_memorysync import MemorySyncCapability

assistant = ConversableAgent("assistant", llm_config=...)

memory = MemorySyncCapability(
    api_key="ms_...",              # or MEMORYSYNC_API_KEY
    user_id="customer-42",         # required — who these memories belong to
    session_id="support-chat",     # scopes the transcript
)
memory.add_to_agent(assistant)     # that's the whole integration

From then on every incoming user message is persisted and enriched with recalled context, and every outgoing reply is persisted — through AG2's own hook system (process_last_received_message + process_message_before_send).

Why this one

Semantics worth knowing

• The reply is never stalled and never broken. Recall blocks at most recall_timeout (default 1.2s); persistence is fire-and-forget off the hot path. Outages and quota exhaustion degrade to "no memories this turn".
• Turns persist under the ag2::<session> session scope with deterministic idempotency seeds — retries and multi-agent echoes converge on one write. Each turn is distilled server-side into durable facts; no raw transcript rows land in the user's memory.
• Roles follow the speaker, not the hook. A message sent by an agent that speaks for a person — a UserProxyAgent, or any agent whose human_input_mode is not "NEVER" — is persisted as a human turn even when the capability is attached to that agent too; every other agent's outgoing message is an ai turn. Only human turns become facts, so this is what makes "attach it to both sides" store the person's words as memory (1.0.2; 1.0.1 labelled them ai when the proxy carried the capability, and they left no memory).
• Tool/function messages are never persisted.
• register_memory_tools(memory, caller=..., executor=...) adds search_memory + save_memory tools (the caller needs an llm_config, as usual for AG2 tools).
• memory.flush() waits for in-flight writes (shutdown/tests); memory.close() flushes and releases the HTTP client.

Configuration

Development

pip install -e . "autogen[openai]" pytest
python -m pytest tests -q      # 25 tests through REAL ConversableAgent chats

The suite includes a reproduction of zep-ag2's documented multi-agent double-store scenario (we store once), a UserProxyAgent + assistant chat with the capability on both sides (the person's words persist as a human turn), and a chat driven from inside asyncio.run() (no deadlock).

License

MIT