mengram-crewai 0.1.0

Mengram memory tools for CrewAI agents — semantic, episodic, and procedural memory (workflows that learn from failures).

mengram-crewai

Mengram memory tools for CrewAI agents — semantic (facts), episodic (events), and procedural memory (workflows that learn from failures).

The unique part is get_procedures: step-by-step playbooks with a success/failure track record and preconditions — the assumptions that were violated when the workflow previously failed — so your agents stop repeating mistakes they already made.

Install

pip install mengram-crewai

Free API key at mengram.io (40 memory adds + 200 searches/mo, no card).

Usage

from crewai import Agent
from mengram_crewai import mengram_tools

agent = Agent(
    role="DevOps engineer",
    goal="Ship the release safely",
    tools=mengram_tools(api_key="om-...", user_id="user-123"),
)

The agent gets three tools:

Tool	What it does
search_memory	Semantic search over facts, preferences, events, decisions
save_memory	Save a durable fact to long-term memory
get_procedures	Learned workflows: steps, track record, preconditions to verify

Multi-user

Pass your end-user's id as user_id — isolated facts, events, workflows, and profile per user under one API key. See mengram.io/for-agents.

Options

mengram_tools(
    api_key="om-...",         # or rely on MENGRAM_API_KEY env var
    user_id="user-123",       # multi-user isolation
    limit=5,                  # max results per tool call
    base_url="https://...",   # self-hosted instances
)

Apache 2.0 · Docs · GitHub