Part V — System and settings
This part covers Chapters 20–25 and gathers system-level settings and maintenance: global settings, theme and appearance, system tray and auto-start, shortcuts and commands, upgrade / backup and sync, and uninstall and data cleanup.
20. Global settings
The Settings panel is organised into 6 groups (regrouped from the old "Account / Desktop / Services" trio — the overloaded Services bucket has been split up; each group caps at 6 tabs):
| Group | Tabs in group | What's inside |
|---|---|---|
| Account | Account, Feedback | Sign-in, usage, plan quota, feedback |
| General | General, Recent, Data, Routines | Theme, tray, auto-start, language, Recent, data backup / cleanup, routines |
| Models | Providers, Model settings, Routing | Provider protocol / credentials / headers, model parameters, per-scenario model routing (main-session scope / subtasks / routines / hooks / sec.decompile / code.ask / memory extraction) |
| Agent | Agents, Reflection, Skills, Tools, Hooks, Context, Memory palace | Personas per mode, the Reflection page (restate intent / auto-reflect / auto-resume tasks), skill loading and signing, tool switches, approval policy, context and compaction, long-term memory |
| Security | Security, Skill/Plugin Signatures, Testing & checks, Data masking, Supply chain / SBOM | Hard guards, signing verification policy and trust store, self-check gate (test / diagnostics / lint / sast / testgen / judge), outbound data-masking engine, SBOM vulnerability-DB sources and offline sync |
| Extensions | Plugins, Messaging | Plugin marketplace (Claude Code compatible), IM channel binding / routing |
See the corresponding chapter for each. The sidebar renders the groups automatically — no component changes.
Settings no longer means "start over when you close it": the Settings panel and the Agent Manager are presented as overlays, so the session panel behind them stays alive and is not destroyed; closing them returns you to the session exactly as it was (scroll position, expanded state and all), and switching back and forth doesn't reload anything.
Two new switches:
| Where | Switch | What it does |
|---|---|---|
| Settings → General | Reveal invisible characters | Marks zero-width / variation-selector and similar invisible characters in the rendered reply, which helps track down "looks identical but won't match" text. Off by default |
| Settings → Memory palace | Enable memory palace | Master switch for the long-term memory subsystem; on by default, and turning it off takes effect with zero restart (see §17.6) |
Invisible-character checking: when suspicious invisible characters are detected in the assistant's thinking or body text, you get a notice saying "N in thinking / N in output" with the characters themselves, and a pointer to turn on "Reveal invisible characters" under Settings → General to see where they are. Detection uses a suspicious-character allowlist — zero-width and variation characters that legitimately occur in Chinese and similar contexts are let through, so it doesn't cry wolf.
[!ref] Cheat sheet: an item-by-item lookup of every setting is in the Reference Manual, "Settings panel".