---
id: "TECH-202610-claude-mem-installer-v15-2-0-silently-uploads-the-last-30-da"
kind: "technique"
aliases: []
title: "claude-mem installer (v15.2.0+) silently uploads the last 30 days of Claude Code, Codex, Gemini CLI and OpenCode prompts to cmem.ai, which passes them to unknown LLM providers"
severity: "medium"
cvss: null
epss: null
exploited: false
status: "open"
category: "ai-tooling"
products: [{"name": "claude-mem", "fixed": [], "affected": ">= 15.2.0", "ecosystem": "npm"}]
ecosystems: []
tags: ["early-warning", "technique"]
not_affected_if: []
published_at: null
first_seen_at: "2026-10-29T14:28:32+00:00"
updated_at: "2026-10-29T14:28:32+00:00"
sources: [{"source": "analyst", "url": null, "first_seen_at": "2026-10-29T14:28:32+00:00"}]
---

# claude-mem installer (v15.2.0+) silently uploads the last 30 days of Claude Code, Codex, Gemini CLI and OpenCode prompts to cmem.ai, which passes them to unknown LLM providers

Running 'npx claude-mem install' interactively with installer v15.2.0 or later (still the case in v15.5.0, released 2026-10-29) opens a browser setup page. As soon as the page loads, it reads the local chat history of the last 30 days from Claude Code, Codex, Gemini CLI, Antigravity CLI and OpenCode, whichever tools you picked. It uploads the full prompt text to cmem.ai along with session IDs, project names, folder paths and timestamps. You cannot stop the upload. The only checkbox is ticked by default and greyed out during the upload, and unticking it does not stop it (confirmed by cmem.ai). cmem.ai says it sent the prompts to a DeepSeek model through OpenRouter, routed to the cheapest provider, without excluding providers that keep or train on data, and it cannot say which provider got them. This contradicts its privacy policy ('nothing leaves your device unless you explicitly turn on cloud sync'). Developers whose prompts contained secrets, internal code, customer data or project names may have leaked them to third parties. The upload is mentioned only in the v15.2.0 changelog (2026-10-25). Reported in GitHub issue #5176. No attacker involved: this is data exposure by the vendor's own installer.

_Summary, check and fix written by a model from the sources; labelled as such._

## Affected

- npm:claude-mem affected >= 15.2.0 fixed ?

## How to check

- Check whether anyone ran 'npx claude-mem install' with installer v15.2.0 or later (from 2026-10-25): run 'npx claude-mem --version' or look at the plugin version in the Claude Code plugins directory.
- Look in proxy, DNS or EDR logs for traffic from developer machines to cmem.ai, especially right after a claude-mem install.
- Check whether the prompts in ~/.claude/projects, ~/.codex/sessions, ~/.gemini and ~/.local/share/opencode from the 30 days before the install contained secrets, credentials or confidential data.

## Fix

- Block cmem.ai on developer machines and networks before anyone runs or reruns the claude-mem installer, or ban claude-mem until the setup upload is opt-in.
- If someone already ran the setup, treat any secrets typed into agent prompts in the 30 days before the install as exposed, and rotate them.
- Ask cmem.ai to delete the install's data. Copies at OpenRouter or the downstream model provider are not covered by that deletion.
- Turn off claude-mem telemetry ('npx claude-mem telemetry disable') and check whether the CMEM Pro or cloud provider is set as the default.

## Timeline

- 2026-10-29T14:28:32+00:00 analyst

## Documents

- [describes] Installing Claude-mem uploads your last 30 days of prompts to cmem.ai, no opt-in https://news.ycombinator.com/item?id=50289137
- [describes] Installing claude-mem uploads your last 30 days of prompts to cmem.ai, no opt-in (issue #5176) https://github.com/thedotmack/claude-mem/issues/5176

---

_Free to read at a human pace. An API with keys and an MCP server are in preparation; early access: hello@infrastack.ai_