crewai-recall

Recall memory for CrewAI agents. Recall learns about the people your crew works for from their conversations, and gives that back as context, search results and answers.

Install

pip install crewai-recall

Create an API key at https://app.recallmem.dev/keys and export it as RECALL_API_KEY (and RECALL_WORKSPACE_ID to pick a workspace). Running your own server? Set RECALL_URL=http://localhost:8000.

Quickstart

from crewai import Agent, Crew, Task
from crewai_recall import recall_context, recall_tools

concierge = Agent(
    role="Travel concierge",
    goal="Plan trips the traveller will love",
    backstory="You remember what each traveller likes.",
    tools=recall_tools("user-42"),  # recall_search, recall_ask, recall_remember
)
plan = Task(
    description="{request}\n\nWhat you know about the traveller:\n{memory}",
    expected_output="A day-by-day plan",
    agent=concierge,
)
request = "Plan a weekend in Lisbon."
Crew(agents=[concierge], tasks=[plan]).kickoff(
    inputs={"request": request, "memory": recall_context("user-42", query=request)}
)

recall_context() reads the person's peer card and the facts that relate to the request (no model call, well under a second), so the crew starts with what Recall knows. The tools let an agent look up more, ask a question of memory and save facts the person states.

What is in the package

The tools work blocking (run) and, on CrewAI versions that run tools asynchronously, with arun. They take user_id (the Recall peer) and an optional client=Recall(...) (or AsyncRecall(...), async only) from recall-mem, and declare RECALL_API_KEY, RECALL_WORKSPACE_ID and RECALL_URL as their environment variables. If Recall cannot be reached, a tool tells the agent so instead of failing the crew.

To have Recall learn from conversations, store them with the SDK, e.g. the request and the crew's answer: recall.session("trip-1").add_messages([user.message(request), crew_peer.message(str(result))]). See the guide.

CrewAI's own memory. CrewAI's Memory takes a pluggable StorageBackend, but that interface is a vector store: CrewAI computes the embeddings and the backend stores and searches vectors. Recall does its own learning and search from text, so it is offered as tools and context rather than as a storage backend.

Docs

• Guide: https://recallmem.dev/guides/crewai
• Recall docs: https://recallmem.dev/docs

MIT licensed.