How Does Mem0 Handle Telemetry and Data Collection? (Opt-Out Guide)

Mem0 ships with built-in, opt-out telemetry that records anonymous usage statistics via PostHog and sends them to the project's analytics endpoint, which can be completely disabled by setting the MEM0_TELEMETRY environment variable to any falsy value before importing the library.

Mem0 (the memory layer for AI applications) includes a telemetry system to help maintainers understand adoption patterns and component usage. According to the mem0ai/mem0 source code, this system is designed to collect only non-identifying metadata while providing a straightforward mechanism to disable all data collection entirely.

How Mem0 Telemetry Works

The telemetry architecture centers around the AnonymousTelemetry class in mem0/memory/telemetry.py. When you import any Mem0 module, the system immediately evaluates whether to initialize the PostHog client or operate in no-op mode based on environment configuration.

The Environment Variable Toggle

Lines 11-18 of mem0/memory/telemetry.py implement the toggle logic:

MEM0_TELEMETRY = os.environ.get("MEM0_TELEMETRY", "True").lower() in ("true", "1", "yes")

The code reads MEM0_TELEMETRY and defaults to "True" if the variable is missing. Strings "true", "1", or "yes" enable telemetry; any other value disables it.

The AnonymousTelemetry Class Constructor

The constructor (lines 25-31) determines the behavior for the entire session. If telemetry is disabled, the class sets self.posthog = None and skips user ID generation. This ensures that all subsequent calls to capture_event() execute an early return without instantiating the PostHog client or generating network traffic.

Global Singleton Instance

Line 60 of mem0/memory/telemetry.py creates a global client_telemetry singleton at module import time:

client_telemetry = AnonymousTelemetry()

This means the telemetry decision happens immediately when the library loads, making pre-import environment configuration critical.

What Data Mem0 Collects (When Enabled)

When telemetry is active, the capture_event() and capture_client_event() functions (lines 73-84) assemble a payload containing only anonymous metadata:

• PostHog project key: phc_hgJkUVJFYtmaJqrvf6CYN67TIQ8yhXAkWzUn9AMU4yX (routes events to the Mem0 project)
• Anonymous user_id: A UUID generated once per installation
• Runtime details: Python version, OS type/version, processor architecture, machine type
• Component configuration: Vector store type, graph store, LLM provider, embedding model, and API version

Notably, the payload excludes user queries, documents, conversation history, or any personal data.

User ID Generation and Storage

The stable identifier is created in mem0/memory/setup.py (lines 35-56) via get_or_create_user_id(). This UUID is stored locally in ~/.mem0/config.json and is never tied to personal information or PII.

How to Disable Mem0 Telemetry

You have three methods to stop data collection, depending on your deployment constraints.

Method 1: Environment Variable (Recommended)

Set MEM0_TELEMETRY before launching Python to prevent the PostHog client from ever being instantiated:

export MEM0_TELEMETRY=False   # "false", "0", or "no" also work

python your_app.py

The variable must be defined before the first import of mem0.memory.telemetry (i.e., before any mem0 modules load).

Method 2: Programmatic Disabling

For CI environments or unit testing, set the variable in Python before importing the library:

import os
os.environ["MEM0_TELEMETRY"] = "False"

import mem0

# Telemetry is now completely disabled

Method 3: Runtime Override

If you must disable telemetry after import (e.g., in a running application), monkey-patch the module flag:

from mem0.memory import telemetry
telemetry.MEM0_TELEMETRY = False

After this assignment, the global client_telemetry singleton continues to exist but all capture_event() and capture_client_event() calls return immediately at the early-exit check (lines 63-66), guaranteeing no network traffic.

Verification and Testing

The test suite in tests/test_telemetry.py (lines 6-16 and 23-31) validates the opt-out behavior. These tests confirm that when MEM0_TELEMETRY is set to False, no PostHog client is created and capture_event() becomes a no-op.

Summary

• Mem0 uses opt-out telemetry via PostHog to collect anonymous usage statistics, defaulting to enabled unless configured otherwise.
• Set MEM0_TELEMETRY=False before importing the library to disable all data collection and prevent PostHog client instantiation.
• When disabled, AnonymousTelemetry sets posthog=None and all capture methods execute immediate returns with zero network overhead.
• Collected data includes only runtime metadata (Python version, OS, component types) and an anonymous UUID, never user content or conversations.
• The anonymous user ID is stored locally in ~/.mem0/config.json and is not linked to personal data.

Frequently Asked Questions

Is Mem0 telemetry enabled by default?

Yes. According to mem0/memory/telemetry.py lines 11-18, if the MEM0_TELEMETRY environment variable is not set, the system defaults to "True" (enabled). The code converts this string to a boolean, meaning telemetry runs unless explicitly disabled.

Can I disable telemetry after importing Mem0?

Yes, though it requires monkey-patching the module. You can set telemetry.MEM0_TELEMETRY = False after importing mem0.memory.telemetry, which triggers the early-exit logic in capture_event(). However, setting the environment variable before import is the only way to guarantee the PostHog client is never instantiated.

Does Mem0 collect conversation content or personal data?

No. The payload construction in mem0/memory/telemetry.py lines 73-84 includes only anonymous metadata such as Python version, operating system details, and component configurations. The system explicitly excludes user queries, document content, and conversation text from all telemetry events.

Where is the anonymous user ID stored?

The UUID is generated once per installation and persisted in a local JSON file at ~/.mem0/config.json, as implemented in mem0/memory/setup.py lines 35-56. Optionally, it may also be stored in the configured vector store metadata, but it remains strictly anonymous and is never linked to personal identifiers or email addresses.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}