Problem

AutoGen agents currently have no built-in mechanism for cross-session memory persistence. Each conversation starts with an empty context, which makes it impossible for agents to recall facts, user preferences, or decisions from prior sessions. Teams building long-running multi-agent workflows need a way to accumulate and retrieve knowledge across executions.

Proposed solution

Add a DakeraMemoryStore integration that gives AutoGen agents access to Dakera — a self-hosted, decay-weighted vector memory server. Dakera provides:

• Semantic search using an on-device ONNX embedding model
• Decay-weighted importance — frequently accessed memories stay relevant; stale facts fade
• Session scoping — memories can be namespaced per agent or session
• Zero external cloud dependencies — runs entirely on-premise via Docker

Usage sketch

import asyncio
from autogen_agentchat.agents import AssistantAgent
from autogen_agentchat.teams import RoundRobinGroupChat
from autogen_ext.memory.dakera import DakeraMemoryStore
from autogen_agentchat.conditions import MaxMessageTermination

store = DakeraMemoryStore(url="http://localhost:3300", api_key="demo", agent_id="support-agent")

agent = AssistantAgent(
    name="support_agent",
    model_client=...,
    memory=[store],
)
team = RoundRobinGroupChat([agent], termination_condition=MaxMessageTermination(5))
asyncio.run(team.run_until_done(task="What has the user complained about before?"))

Self-host Dakera:

docker run -p 3300:3300 -e DAKERA_API_KEY=demo ghcr.io/dakera-ai/dakera:latest

PR #7900 contains the implementation (in python/packages/autogen-ext/src/autogen_ext/memory/dakera/).