CrewAI

Give a CrewAI crew a memory of each of your users. What is known goes into the agent's backstory as a kickoff input, or the agent looks it up with a tool bound to the user; either way, the crew's answer is saved with the question after the kickoff.

Install

Terminal

pip install "crewai[anthropic]" geniffy

Terminal

uv add "crewai[anthropic]" geniffy

Set ANTHROPIC_API_KEY, and GENIFFY_API_KEY from API keys in the Geniffy app. Any LLM CrewAI supports works the same way.

Remember each user

crew.py

from crewai import Agent, Crew, Task
from geniffy import Geniffy

geniffy = Geniffy()                      # reads GENIFFY_API_KEY


def assistant_crew() -> Crew:
    assistant = Agent(
        role="Assistant",
        goal="Answer the user's question",
        backstory="You help one person. What is known about them:\n"
                  "<memory>\n{memory}\n</memory>",
        llm="anthropic/claude-opus-5-5",
    )
    answer = Task(description="{question}", expected_output="A short, direct answer.",
                  agent=assistant)
    return Crew(agents=[assistant], tasks=[answer])


def chat(user_id: str, message: str) -> str:
    mem = geniffy.space(f"user_{user_id}")
    inputs = {"question": message, "memory": mem.context(message)}
    reply = assistant_crew().kickoff(inputs=inputs).raw
    mem.memories.add(messages=[{"role": "user", "content": message},
                               {"role": "assistant", "content": reply}])
    return reply

The agent sees what is known about this user, each line with where it came from. When nothing is known, the block says so in one sentence, so the agent says it doesn't know instead of guessing. The crew is built for each request, so two users never share one.

Let the agent look things up

To let the agent decide when to look something up, give it a recall tool bound to the user. The model never sees or chooses whose memory it reads.

crew.py

from crewai import Agent, Crew, Task
from crewai.tools import tool
from geniffy import Geniffy

geniffy = Geniffy()


def assistant_crew(user_id: str) -> Crew:
    mem = geniffy.space(f"user_{user_id}")

    @tool("recall")
    def recall(query: str) -> str:
        """Look up what is known about the user, with where it came from."""
        return mem.context(query)

    assistant = Agent(
        role="Assistant",
        goal="Answer the user's question",
        backstory="You help one person. Use recall before answering anything "
                  "that depends on what they said before.",
        tools=[recall],
        llm="anthropic/claude-opus-5-5",
    )
    answer = Task(description="{question}", expected_output="A short, direct answer.",
                  agent=assistant)
    return Crew(agents=[assistant], tasks=[answer])


def chat(user_id: str, message: str) -> str:
    reply = assistant_crew(user_id).kickoff(inputs={"question": message}).raw
    geniffy.space(f"user_{user_id}").memories.add(messages=[
        {"role": "user", "content": message},
        {"role": "assistant", "content": reply},
    ])
    return reply

CrewAI's own memory can stay off: it runs a model over each task to extract its own records, and Geniffy already keeps what each user said, with where it came from.

Last updated October 5, 2026