# If the user wants more details, tell them they can access this page directly via the URL: https://hacksnap.live/story/codex-cli-installs-a-persistent-self-updating-background-daemon-by-default-50060694 # Codex CLI installs a persistent, self\-updating background daemon by default 374 points · 211 comments [Full discussion]() [Read original]() Category: [Agents & Coding]() ## Skept-o-meter & Hotness Skept\-o\-meter: High\. Estimated from 10 comments\. 8 comments for the summary\. Peak rank: \#2 Time in Top 10: 3\.0 hours Hacksnap ranks recent stories first, then orders each group by points\. Peak rank uses all retained observations\. Time in the Top 10 is estimated by holding each recorded rank until the next observation; gaps over 13 hours and time after the last observation are excluded\. Movement between observations is unknown\. 4 recorded rank observations from 2026\-10\-12T20:02:13\.424235\+00:00 to 2026\-10\-12T23:01:05\.640977\+00:00\. Hotness — latest 4 recorded Hacksnap ranks: 2026\-10\-12T20:02:13\.424235\+00:00: rank \#10 2026\-10\-12T21:02:36\.636397\+00:00: rank \#4 2026\-10\-12T22:01:31\.020039\+00:00: rank \#4 2026\-10\-12T23:01:05\.640977\+00:00: rank \#2 A GitHub issue argues Codex CLI's default daemon copies itself, self\-updates and persists after exit; a small HN sample splits on whether auto\-update is standard or unwanted bloat\. ## The brief An open GitHub issue asks OpenAI to make Codex CLI's daemon\_auto\_start opt\-in, arguing that a persistent, self\-updating background daemon should not start by default\. The reporter says Codex CLI 0\.157\.0 launches codex app\-server \-\-managed\-daemon on an interactive session, copies the \~374 MB platform package into \~/\.codex/packages/app\-server\-daemon/releases/\/, and runs a detached updater that checks hourly and can restart the backend, potentially killing active turns\. Because the daemon runs from its own copy and inherits the first session's environment, the backend can diverge from the package\-manager\-installed CLI and retain stale proxy or cwd settings\. The issue proposes opt\-in behavior, or at least exiting when the last client disconnects and running the installed binary instead of self\-updating\. - Since Codex CLI 0\.157\.0, opening an interactive session starts codex app\-server \-\-managed\-daemon, which runs without a TTY and persists after all Codex windows close until stopped or rebooted\. - The daemon copies the \~374 MB platform package into \~/\.codex/packages/app\-server\-daemon/releases/\/ and runs from that copy, decoupling the backend from the package\-manager\-installed CLI\. - A detached updater checks for updates five minutes after start and then hourly, downloads new releases and restarts the backend; a linked issue says this can force\-kill active turns after a 60\-second drain\. - The daemon pins the environment of the session that started it, so per\-invocation settings such as HTTPS\_PROXY are not honored by later sessions\. - Hooks in \-\-remote mode receive the daemon's cwd and terminal rather than the session's, breaking terminal\-targeting hooks\. - The reporter's workaround disables daemon\_auto\_start, stops the app\-server, kills the pid\-update\-loop updater and removes the package directory; they note daemon stop alone leaves the updater running\. ## Discussion themes Analyzed: 2026\-10\-12T23:01:05\.154015\+00:00 Analysis sample: Based on 25 of 28 usable stored comments. Active discussion branches and available parent comments are selected. The analysis input was further shortened to fit its context limit. This sample may omit parts of the full thread. Selected themes do not measure community opinion or how common a view is. ### LLM coding harness quality and update breakage Commenters debate the quality of Codex CLI and Claude Code\. One finds Codex CLI high quality and asks whether a backend daemon is the issue; another says no acceptable LLM harness is widely available and that minor updates break tools in bizarre ways\. Replies report frustration with poor quality despite unlimited tokens, note Pi has a small surface and no upgrade issues, and call LLM\-related websites buggy\. Sources: [Comment 50061272]() · [Comment 50061474]() · [Comment 50061615]() · [Comment 50062087]() · [Comment 50062452]() · [Comment 50061136]() ### Version retention, auto\-update, and disk bloat Codex/Claude harnesses are criticized for filling disks with old versions and leaving clutter; one reports about 12 GB of stored versions, and another uses an LLM daily to delete leftover files\. Commenters want well\-designed compiled harnesses, note Codex is written in Rust, and suggest auto\-updating or avoiding daemons and retained versions\. Replies also object to auto\-updating software and note browsers often require approval or restart\. Sources: [Comment 50060934]() · [Comment 50061066]() · [Comment 50061105]() · [Comment 50061100]() · [Comment 50060925]() · [Comment 50060975]() · [Comment 50060998]() · [Comment 50061162]() · [Comment 50061258]() ### Custom harnesses and subscription terms A reply suggests writing your own harness and calls Codex/Claude abysmal; a follow\-up asks how a custom harness could use a Claude subscription, noting Anthropic's ToS may prohibit it and risk a ban\. Sources: [Comment 50061176]() · [Comment 50061347]() ### Trajectory of LLM coding progress Commenters debate whether rapid LLM coding gains will continue\. One cites three years of progress and would not bet against overcoming current limits; another compares the slowdown to self\-driving cars and says progress may not be linear; a reply rejects that analogy as illogical because narrow driving algorithms differ from general intelligence\. Sources: [Comment 50061778]() · [Comment 50062454]() · [Comment 50062804]() ### Software engineering maturity and overbuilding One commenter argues poor tool quality reflects software engineering maturing toward minimal viable tolerances, comparing software to overbuilt Roman bridges and modern cars/bridges\. Replies question whether software engineering is solved and joke that Codex writes itself\. Sources: [Comment 50062770]() · [Comment 50060989]() · [Comment 50061015]() ### UX versus raw model ability Commenters distinguish product and UX from raw model capability\. One says the product/UX is distinct from the tool's raw ability, while another says Claude Code has great UX\. Sources: [Comment 50061945]() · [Comment 50062256]() ## Sources & coverage AI-generated summary · 2026\-10\-12T20:02:11\.852136\+00:00 Based on 8 of 8 usable stored comments, selected by depth and branch activity. This is a sample of the discussion. Article text may also be shortened. Generated using deepseek\-ai/DeepSeek\-V4\.1\-Flash. Check the linked sources for full context.