---
id: "lib-2026-09-25-codex-cli-make-daemon-auto-start-opt-in-48195"
type: "library"
title: "Codex CLI: make daemon_auto_start opt-in (#48195)"
lang: "EN"
org: "GitHub (openai/codex)"
source_type: "Practitioner post"
date: "2026-09-25"
week: "2026-W42"
url: "https://nosov.ai/library/lib-2026-09-25-codex-cli-make-daemon-auto-start-opt-in-48195/"
original_url: "https://github.com/openai/codex/issues/48195"
tier: "B"
tier_name: "Useful, limited"
tier_reason: "A user-filed GitHub issue on OpenAI's Codex repository with exact process names, paths and a reproducible workaround, confirmed by other users; first-hand and specific, but one report with no vendor response in the thread."
summary_basis: "full-text"
original_minutes: "8"
tags: ["coding agents", "endpoint governance"]
---

# Codex CLI: make daemon_auto_start opt-in (#48195)

Tier B (Useful, limited): A user-filed GitHub issue on OpenAI's Codex repository with exact process names, paths and a reproducible workaround, confirmed by other users; first-hand and specific, but one report with no vendor response in the thread.

Summary based on the full text.

## In short

GitHub issue #48195 on openai/codex, opened 25 September 2026. Since version 0.157.0, opening an interactive Codex CLI session starts a background server that keeps running after every window closes, runs from its own copied binaries, updates itself hourly and can restart mid-task. The reporter asks for the feature to be opt-in for the CLI. The thread drew wide agreement and attention on Hacker News around 12 October; at capture the issue was still open with no maintainer reply.

## Key numbers

- 0.157.0: Codex CLI version that began starting a persistent background daemon by default
- ~374 MB: platform package the CLI copies into a managed release folder to run the daemon
- 5 minutes: delay before the detached updater first checks for releases, then hourly
- 1.3GB: background updates one commenter found downloaded a few days after install

## What it reports

- Starting `codex` launches `codex app-server --managed-daemon` with no terminal, reparented to PID 1, alive until stopped or reboot.
- The CLI copies about 374 MB of binaries (including bundled bwrap, zsh and rg) into `~/.codex/packages/app-server-daemon/`, so the running backend can drift from the version the package manager installed.
- A detached updater checks 5 minutes after start and then hourly, downloads releases and restarts the backend; a linked issue says this can kill active turns after a 60-second drain.
- The daemon keeps the environment of the first session that started it, including proxy variables, so later sessions' outbound traffic can go through the wrong proxy.
- `codex app-server daemon stop` leaves the updater process running.
- Commenters report the CLI unusable on Windows, one finds no `daemon_auto_start` entry in the official config reference, and one says the 0.157.0 release note gave the change a single sentence.

## How they got it

The reporter observed process trees, file paths and network behaviour on Linux (WSL2) with an npm install and documented a workaround: disable the feature flag, stop the daemon, kill the updater, delete the copied package. Others confirmed on Windows 10 and Arch Linux packaging.

## Limits

- User reports, not a security audit; no response from OpenAI appears in the captured thread.
- Several comments are opinion ("malware") rather than evidence.
- Behaviour may change in later releases; the capture reflects versions 0.157.0-0.160.x.

## Why it matters for companies

Coding agents installed on staff machines can start persistent, self-updating processes that IT did not approve and that run code outside the package manager's control. That breaks software inventory, change control and proxy policy at once. Firms rolling out agent CLIs should check for background services and auto-updaters, pin configuration centrally (here `daemon_auto_start = false`), and read release notes as change requests. One commenter said they were withdrawing their company's licences, which shows how fast trust erodes when defaults change silently.

Original: https://github.com/openai/codex/issues/48195