OpenAI Codex 0.160 ships projectless sessions and Guardian recall
The CLI's latest build keeps subagents' environments alive, makes GPT-6.1 Sol the default model, and clears SQLite stalls plus Windows sandbox snags.
Copy markdown
Sessions can start outside a project
Launch Codex anywhere and it now falls back to workspace defaults instead of demanding a project first, restoring permissions when you resume (PR #49160) — as long as your org's policy allows it. Quick scratch work and one-off repos stop fighting the setup step.
Guardian review that remembers what you told it
An opt-in Guardian review can now pull back your earlier instructions and agent-handoff context during a pass (PRs #49036, #49057), so it stops re-flagging things you already resolved. Less noise on long agent runs.
Subagents keep their environment on startup
Subagents that are still starting now hold onto their environment and surface configuration or preparation-failure notices instead of dying silently (PR #49075). Fewer mystery failures when you fan work out across agents.
GPT-6.1 Sol is the default model now
Since 0.159.1, Codex bundles GPT-6.1 Sol as the default in its own catalog and in the Amazon Bedrock Mantle and Runtime catalogs. That's near-Astra coding at roughly a fifth of Astra's API price, with cached input cut to $0.10 per million tokens — now what you get out of the box.
SQLite and the Windows sandbox get unstuck
0.160 prevents SQLite stalls during connection setup and logging (surfacing real init errors instead of fake timeouts), fixes Windows PowerShell sandbox fallbacks and long-path permission repairs, and reclaims unused log-database space in the background. Less hanging, smaller footprint.