AG2 Integration

CortexDB provides persistent long-term memory for AG2 agents, enabling multi-agent conversations that remember context across sessions. AG2 is the successor to Microsoft AutoGen, featuring an event-driven architecture with improved modularity.

Installation

pip install cortexdb[ag2]

Quick Start

from ag2 import AssistantAgent, UserProxyAgent
from cortexdb import Cortex
from cortexdb_ag2 import CortexDBAgent

client = Cortex(base_url="http://localhost:3141", api_key="your-cortex-api-key")

# Create a memory-augmented assistant
agent = CortexDBAgent(
    name="assistant",
    cortex_client=client,
    tenant_id="my-app",
    namespace="ag2-assistant",
    memory_top_k=10,
    llm_config={"model": "gpt-4o"},
    system_message="You are a helpful assistant with long-term memory.",
)

user = UserProxyAgent(
    name="user",
    human_input_mode="ALWAYS",
)

# Conversations are automatically stored and recalled
user.initiate_chat(agent, message="What do you remember about our project?")

As Memory Backend

The CortexDBAgent extends AG2's ConversableAgent with automatic memory:

• Auto-recall: Before each response, relevant memories are retrieved and injected as context
• Auto-store: After each response, the conversation turn is persisted to CortexDB

# Disable auto behaviors for manual control
agent = CortexDBAgent(
    name="assistant",
    cortex_client=client,
    tenant_id="my-app",
    auto_store=False,
    auto_recall=False,
    llm_config={"model": "gpt-4o"},
)

As Agent Tools

Register CortexDB as callable tools that AG2 agents can invoke:

from ag2 import ConversableAgent, UserProxyAgent
from cortexdb import Cortex
from cortexdb_ag2 import register_cortexdb_tools

client = Cortex(base_url="http://localhost:3141", api_key="your-cortex-api-key")

assistant = ConversableAgent("assistant", llm_config={"model": "gpt-4o"})
user_proxy = UserProxyAgent("user_proxy")

# Register search, store, and forget tools
register_cortexdb_tools(assistant, user_proxy, client, tenant_id="my-app")

user_proxy.initiate_chat(
    assistant,
    message="Search your memory for what we discussed about deployment.",
)

You can also register tools individually:

from cortexdb_ag2 import cortexdb_search_fn, cortexdb_store_fn

search = cortexdb_search_fn(client, tenant_id="my-app")
store = cortexdb_store_fn(client, tenant_id="my-app")

assistant.register_for_llm(description="Search long-term memory")(search)
user_proxy.register_for_execution()(search)

Configuration

| Parameter | Default | Description | |---|---|---| | base_url | http://localhost:3141 | CortexDB server URL | | api_key | None | CortexDB API key | | tenant_id | "default" | Tenant identifier | | namespace | None | Memory namespace | | memory_top_k | 5 | Results per recall | | auto_store | True | Auto-store conversation turns | | auto_recall | True | Auto-retrieve context before responding |

Complete Example

from ag2 import UserProxyAgent
from cortexdb import Cortex
from cortexdb_ag2 import CortexDBAgent, register_cortexdb_tools

client = Cortex(base_url="http://localhost:3141", api_key="your-cortex-api-key")

# Memory-augmented agent with tool access
agent = CortexDBAgent(
    name="knowledge_agent",
    cortex_client=client,
    tenant_id="engineering",
    namespace="project-alpha",
    memory_top_k=10,
    llm_config={"model": "gpt-4o"},
    system_message=(
        "You are an engineering assistant with long-term memory. "
        "You remember past conversations and decisions."
    ),
)

user = UserProxyAgent(name="engineer", human_input_mode="ALWAYS")

# Also register explicit tools for manual memory operations
register_cortexdb_tools(agent, user, client, tenant_id="engineering")

# Start conversation — memories persist across sessions
user.initiate_chat(
    agent,
    message="Remember: we decided to use PostgreSQL for the user service.",
)

# In a later session, the agent recalls this decision automatically
user.initiate_chat(
    agent,
    message="What database did we choose for the user service?",
)