plugin-autogen

plugin-autogen

Neocortex (TinyHuman) memory tools for Microsoft AutoGen.

Features

Provides explicit Python functions that allow AutoGen agents to use a Mastra-aligned Neocortex tool surface.

• NeocortexMemoryTools: Class wrapping the memory client and exposing the tool methods.
• register_neocortex_tools: Helper function to register the tools to AutoGen agents.

Tool methods:

• save_memory, recall_memory, delete_memory
• sync_memory
• insert_document, insert_documents_batch
• list_documents, get_document, delete_document
• query_memory_context, chat_memory_context
• record_interactions, recall_thoughts
• chat_memory, interact_memory
• recall_memory_master, recall_memories
• get_ingestion_job

Document Insert Contract

• insert_document requires document_id.
• insert_documents_batch requires document_id on every item in items_json.

Installation

pip install neocortex-autogen

Usage

Set your API key:

export TINYHUMANS_API_KEY="your_token_here"

Expose tools to your AutoGen agent

NeocortexMemoryTools(...) only creates Python methods.
To make them callable by the agent, you must register them with register_neocortex_tools(...).

AutoGen uses a caller/executor split:

• caller: usually AssistantAgent (LLM proposes tool calls)
• executor: usually UserProxyAgent (runs tool functions locally)

If you skip registration, the agent cannot see or call any Neocortex tools.

Instantiate the tools with a TinyHumanMemoryClient and register them to your agents.

import os
import autogen
from tinyhumansai import TinyHumanMemoryClient
from neocortex_autogen import NeocortexMemoryTools, register_neocortex_tools

client = TinyHumanMemoryClient(token=os.getenv("TINYHUMANS_API_KEY"))
memory_tools = NeocortexMemoryTools(client=client, default_namespace="autogen_memory")

assistant = autogen.AssistantAgent("assistant", llm_config={"config_list": [...]})
user_proxy = autogen.UserProxyAgent("user_proxy", human_input_mode="NEVER")

# The assistant proposes tools, the user_proxy executes them locally.
register_neocortex_tools(tools=memory_tools, caller=assistant, executor=user_proxy)

user_proxy.initiate_chat(assistant, message="Remember my name is Alice.")

OpenCode-style prompting for tool use

After registration, add clear instructions so your agent actually uses the tools:

assistant = autogen.AssistantAgent(
    "assistant",
    llm_config={"config_list": [...]},
    system_message=(
        "You have Neocortex memory tools. "
        "Use save_memory for new facts, recall_memory/query_memory_context before answering "
        "memory questions, and delete_memory only when explicitly requested."
    ),
)

E2E example

Run the package smoke example:

cd packages/plugin-autogen
python3 examples/autogen_agent.py

It exercises registration and tool-calling across save/recall, documents, context/chat, interactions, sync, ingestion status, and cleanup.

Available registered tools

register_neocortex_tools(...) registers:

• save_memory, recall_memory, delete_memory
• sync_memory
• insert_document, insert_documents_batch
• list_documents, get_document, delete_document
• query_memory_context, chat_memory_context
• record_interactions, recall_thoughts
• chat_memory, interact_memory
• recall_memory_master, recall_memories
• get_ingestion_job