Jason Liu outlines a personal operating system built on Codex: durable pinned threads, compaction, Obsidian vault memory, Heartbeats, and Goals for long-running knowledge work.
Key Takeaways
Durable threads + compaction: pinned megathreads (Chief of Staff, Agents SDK, open source) accumulate history across sessions; compaction keeps them alive without full message replay, at higher cost.
Vault-as-memory: an Obsidian vault tracked in GitHub serves as shared external memory; agents write diffs so you can review what the model decided to remember.
Heartbeats: thread-local recurring schedules let Codex monitor Slack, Gmail, PRs, or Google Docs comments on a cadence and draft replies or re-render artifacts without human presence.
Computer/browser/chrome split: $browser for local apps, @chrome for multi-tab authenticated sessions, @computer for GUI-only tasks; connectors ($slack, $gmail, $calendar) bridge to where real work lives.
Goals need oracles: a strong Goal requires a verifiable success criterion – Liu used the Python Rich library’s existing unit test suite as the oracle for a Rust migration.
Hacker News Comment Review
Heavy skepticism that this workflow produces measurable outcomes; the core challenge is that the post describes process elaboration with no shipped artifacts shown as evidence.
Commenters flagged a conflict of interest: the author works on the Codex team at OpenAI, framing the post as marketing dressed as personal workflow.
A practical bug undermines the pitch: long Codex chats already freeze the UI within a day, directly contradicting the durable-megathread premise.
Notable Comments
@4k0hz: author works at OpenAI on the Codex team – undisclosed in the post.
@manuisin: Codex UI freezes and lags before long chats even reach one day old, breaking the durable-thread model in practice.
