mem0ai/mem0 · critical · ImportError

Could not import faiss python package. Please install it wit

Error message

Could not import faiss python package. Please install it with `pip install faiss-gpu` (for CUDA supported GPU) or `pip install faiss-cpu` (depending on Python version).

What it means

ImportError raised at import time when the FAISS C extension cannot be imported. faiss.py first silences SWIG-related DeprecationWarnings, then imports faiss; if the package is absent (or its native .so fails to load), the ImportError is re-raised with guidance to install faiss-cpu or faiss-gpu. The SafeUnpickler machinery further down never runs — this is purely an environment/dependency failure.

Source

Thrown at mem0/vector_stores/faiss.py:23

import uuid
import warnings
from pathlib import Path
from typing import Any, Dict, List, Optional

import numpy as np
from pydantic import BaseModel

try:
    # Suppress SWIG deprecation warnings from FAISS
    warnings.filterwarnings("ignore", category=DeprecationWarning, message=".*SwigPy.*")
    warnings.filterwarnings("ignore", category=DeprecationWarning, message=".*swigvarlink.*")

    logging.getLogger("faiss").setLevel(logging.WARNING)
    logging.getLogger("faiss.loader").setLevel(logging.WARNING)

    import faiss
except ImportError:
    raise ImportError(
        "Could not import faiss python package. "
        "Please install it with `pip install faiss-gpu` (for CUDA supported GPU) "
        "or `pip install faiss-cpu` (depending on Python version)."
    )

from mem0.vector_stores.base import VectorStoreBase

logger = logging.getLogger(__name__)


class SafeUnpickler(pickle.Unpickler):
    """
    Restricted unpickler that only allows safe built-in types.

    This prevents arbitrary code execution via pickle deserialization by only
    allowing a whitelist of safe types (dict, list, str, int, float, bool, tuple, None).
    """

View on GitHub (pinned to 001c235229)

Solutions

1. Install the right flavor: pip install faiss-cpu (no GPU) or pip install faiss-gpu (CUDA machine).
2. If import still fails after install, install the missing native runtime (e.g. apt-get install libgomp1) or switch to a manylinux-based image.
3. Match the faiss wheel to your Python version (3.9-3.12 supported by mem0).

Example fix

# before
from mem0.vector_stores.faiss import FaissDB  # ImportError

# after
# pip install faiss-cpu
from mem0.vector_stores.faiss import FaissDB

try:
    import faiss  # noqa: F401
    FAISS_OK = True
except ImportError:
    FAISS_OK = False

if provider == "faiss" and not FAISS_OK:
    raise SystemExit("faiss missing: pip install faiss-cpu (or faiss-gpu)")

When it happens

Trigger: Importing mem0.vector_stores.faiss without faiss installed; a broken install where the wheel's native library fails to load against the system's libgomp/OpenBLAS; installing faiss-gpu on a machine without CUDA and having the import fail.

Common situations: Local experiments defaulting to the FAISS backend on a fresh venv; Alpine/slim Docker images missing OpenMP libraries needed by faiss-cpu; installing faiss-cpu on old Python versions with no matching wheel.

Related errors

• The 'upstash_vector' library is required. Please install it
• The 'boto3' library is required. Please install it using 'pi
• FastEmbed is not installed. Please install it using `pip in
• langchain is not installed. Please install it using `pip ins
• The 'ollama' library is required. Please install it using 'p

AI-assisted analysis of mem0ai/mem0@001c235229 (2026-08-15). Data as JSON: /api/errors/0bfc4d1dbcab183b. Report an issue: GitHub.