AutoGen Integration¶

Engramia provides EngramiaMemory — an implementation of AutoGen's Memory interface that plugs directly into AssistantAgent.

Installation¶

pip install "engramia[openai,autogen]"

Quick start¶

from autogen_agentchat.agents import AssistantAgent
from autogen_ext.models.openai import OpenAIChatCompletionClient
from engramia import Memory
from engramia.providers import OpenAIProvider, OpenAIEmbeddings, JSONStorage
from engramia.sdk.autogen import EngramiaMemory, learn_from_result

mem = Memory(
    llm=OpenAIProvider(model="gpt-4.1"),
    embeddings=OpenAIEmbeddings(),
    storage=JSONStorage(path="./engramia_data"),
)

model_client = OpenAIChatCompletionClient(model="gpt-4o")
agent = AssistantAgent(
    name="coder",
    model_client=model_client,
    system_message="You are a senior Python developer.",
    memory=[EngramiaMemory(mem)],
)

result = await agent.run(task="Build a CSV parser")

# Learn from the result (AutoGen Memory has no post-run hook)
learn_from_result(mem, task="Build a CSV parser", result=result)

What happens behind the scenes:

Before each LLM call, update_context() recalls relevant patterns and injects a SystemMessage
Agent executes normally with the enriched context
After the run, call learn_from_result() to store the output

EngramiaMemory parameters¶

Parameter	Default	Description
`recall_limit`	`3`	Max patterns to recall per LLM call
`name`	`"engramia"`	Display name for this memory source

learn_from_result parameters¶

Parameter	Default	Description
`task`	—	Task description (required)
`result`	—	AutoGen `TaskResult` from `agent.run()`
`eval_score`	`7.0`	Eval score to assign to the learned pattern

With teams¶

from autogen_agentchat.teams import RoundRobinGroupChat
from autogen_agentchat.conditions import MaxMessageTermination

coder = AssistantAgent(
    name="coder",
    model_client=model_client,
    memory=[EngramiaMemory(mem)],
)
reviewer = AssistantAgent(
    name="reviewer",
    model_client=model_client,
    system_message="Review code for quality and suggest improvements.",
    memory=[EngramiaMemory(mem)],
)

team = RoundRobinGroupChat(
    [coder, reviewer],
    termination_condition=MaxMessageTermination(6),
)

result = await team.run(task="Build and review a CSV parser")
learn_from_result(mem, task="Build and review a CSV parser", result=result)

Direct query and add¶

ag_mem = EngramiaMemory(mem)

# Query patterns directly
result = await ag_mem.query("CSV parsing")
for item in result.results:
    print(item.content)

# Add patterns manually
from autogen_core.memory import MemoryContent
await ag_mem.add(MemoryContent(content="Use pandas for CSV: pd.read_csv('file.csv')"))

Multiple memory sources¶

AutoGen supports multiple memory instances — combine Engramia with other sources:

from autogen_core.memory import ListMemory

local_notes = ListMemory(name="notes")
await local_notes.add(MemoryContent(content="Always validate input before processing."))

agent = AssistantAgent(
    name="coder",
    model_client=model_client,
    memory=[EngramiaMemory(mem), local_notes],
)