autogen-dakera

Persistent, semantically-recalled memory for AutoGen agents, powered by Dakera.

Your AutoGen agents remember everything — across sessions, across restarts. Dakera handles embedding, storage, and retrieval server-side with no local model required.

---

Quick Start

Step 1 — Run Dakera

Dakera is a self-hosted memory server. Spin it up with Docker:

docker run -d \
  --name dakera \
  -p 3300:3300 \
  -e DAKERA_ROOT_API_KEY=dk-mykey \
  ghcr.io/dakera-ai/dakera:latest

For a production setup with persistent storage, use Docker Compose:

# Download and start
curl -sSfL https://raw.githubusercontent.com/Dakera-AI/dakera-deploy/main/docker-compose.yml \
  -o docker-compose.yml
DAKERA_API_KEY=dk-mykey docker compose up -d

# Verify it's running
curl http://localhost:3300/health

Full deployment guide: github.com/Dakera-AI/dakera-deploy

Step 2 — Install the integration

pip install autogen-dakera

Step 3 — Add memory to your agent

from autogen_agentchat.agents import AssistantAgent
from autogen_agentchat.teams import RoundRobinGroupChat
from autogen_agentchat.conditions import MaxMessageTermination
from autogen_ext.models.openai import OpenAIChatCompletionClient
from autogen_dakera import DakeraMemory

memory = DakeraMemory(
    api_url="http://localhost:3300",
    api_key="dk-mykey",
    agent_id="my-agent",
)

model_client = OpenAIChatCompletionClient(model="gpt-4o")

agent = AssistantAgent(
    name="assistant",
    model_client=model_client,
    memory=[memory],
)

# Agent now persists what it learns across sessions

---

Installation

# Core + integration
pip install autogen-dakera

# With AutoGen (if not already installed)
pip install "autogen-dakera[autogen]"

Requirements: Python ≥ 3.10, a running Dakera server (see Step 1 above)

---

Configuration

Parameter	Type	Default	Description
api_url	str	—	Dakera server URL (e.g. http://localhost:3300)
api_key	str	""	API key set via DAKERA_ROOT_API_KEY
agent_id	str	—	Logical identifier for this agent's memory
min_importance	float	0.0	Minimum importance score for recalled memories
top_k	int	5	Number of memories to surface per query

---

Examples

Multi-agent team with shared memory

import asyncio
from autogen_agentchat.agents import AssistantAgent
from autogen_agentchat.teams import RoundRobinGroupChat
from autogen_agentchat.conditions import MaxMessageTermination
from autogen_ext.models.openai import OpenAIChatCompletionClient
from autogen_dakera import DakeraMemory

async def main():
    shared_memory = DakeraMemory(
        api_url="http://localhost:3300",
        api_key="dk-mykey",
        agent_id="research-team",
        top_k=8,
    )

    model_client = OpenAIChatCompletionClient(model="gpt-4o")

    researcher = AssistantAgent(
        name="researcher",
        model_client=model_client,
        memory=[shared_memory],
        system_message="You are a research expert. Remember key findings.",
    )

    analyst = AssistantAgent(
        name="analyst",
        model_client=model_client,
        memory=[shared_memory],
        system_message="You are a data analyst. Build on what the researcher found.",
    )

    team = RoundRobinGroupChat(
        [researcher, analyst],
        termination_condition=MaxMessageTermination(max_messages=6),
    )

    # First session — agents learn and store
    result = await team.run(task="Research AI memory architectures")
    print(result.messages[-1].content)

    # Later session — agents recall prior research
    result = await team.run(task="What do we know about transformer memory?")
    print(result.messages[-1].content)

asyncio.run(main())

Filtering memories by importance

memory = DakeraMemory(
    api_url="http://localhost:3300",
    api_key="dk-mykey",
    agent_id="my-agent",
    min_importance=0.7,  # only surface high-quality memories
    top_k=3,
)

---

How it works

1. During conversation, AutoGen calls DakeraMemory.add() with new messages
2. Dakera embeds the content server-side and stores it with a semantic vector
3. Before each agent response, AutoGen calls DakeraMemory.query() — Dakera performs hybrid search and returns the most relevant past memories
4. Memories are injected into the agent's context automatically

---

Related packages

Package	Framework	Language
crewai-dakera	CrewAI	Python
langchain-dakera	LangChain	Python
llamaindex-dakera	LlamaIndex	Python
@dakera-ai/langchain	LangChain.js	TypeScript

---

Links

• Dakera Server — self-hosted memory server
• Dakera Python SDK — low-level API client
• Integration guide — full setup walkthrough
• All integrations

---

License

MIT © Dakera AI

---

dakera.ai · Documentation · Request Early Access