AutoGen / AG2 integration

Astrocyte memory for Microsoft AutoGen multi-agent conversations, with both direct API and tool registration.

Module: astrocyte.integrations.autogen Pattern: Memory API (save/query/get_context) + OpenAI-format tool export Framework dependency: ag2 (optional)

Install

pip install astrocyte ag2

Usage — Direct memory API

from astrocyte import Astrocyte
from astrocyte.integrations.autogen import AstrocyteAutoGenMemory

brain = Astrocyte.from_config("astrocyte.yaml")
memory = AstrocyteAutoGenMemory(
    brain,
    bank_id="team",
    agent_banks={"agent-a": "bank-a", "agent-b": "bank-b"},
)

# Save in conversation hooks
await memory.save("User prefers Python", agent_id="agent-a")

# Get formatted context for injection
context = await memory.get_context("What does the user prefer?")
# → "- User prefers Python"

Usage — Tool registration

# Export as OpenAI-format tools for function calling
tools = memory.as_tools()
handlers = memory.get_handlers()

# Register with ConversableAgent
agent = ConversableAgent("assistant", llm_config=llm_config)

Integration pattern

Method	Astrocyte call
save(content, agent_id=...)	brain.retain() with agent metadata
query(query, agent_id=...)	brain.recall() → list of hit dicts
get_context(query)	brain.recall() → formatted string
as_tools()	OpenAI-format tool definitions
get_handlers()	Async handler functions for tool dispatch

API reference

AstrocyteAutoGenMemory(brain, bank_id, *, agent_banks=None)

Parameter	Type	Description
brain	Astrocyte	Configured Astrocyte instance
bank_id	str	Shared team bank
agent_banks	dict[str, str]	Agent ID → bank ID for per-agent memory