Settings
The settings tabs, the defaults that matter, and the two that change your safety posture.
Settings opens with Ctrl , (⌘ , on macOS). The tabs are grouped under You & workspace, Look & feel, Agents, Voice, Connect, and About. Most are ordinary preferences. Two settings are not, and they are covered first.
The two that matter for safety
YOLO defaults to on
Found under Agents / Harmony. It injects provider full-access flags at launch.
WARNING
This defaults to on. It removes friction and it removes a safety rail at the same time. Before pointing agents at a repository you care about, open this tab and decide deliberately rather than inheriting the default.
YOLO is an authority setting. It is separate from the per-session /mode, and
separate again from Crew Mode (Experimental), which governs delegation scale. Three switches;
turning one on does not turn the others on.
A2A mode
Also under Agents / Harmony. Off, Connect, or Orchestrate — whether agents may
message and coordinate with each other at all.
If you do not want agents talking, Off is more reliable than hoping they stay
on topic.
The tabs
| Tab | Covers |
|---|---|
| General | MCP host registration per target, update channel, experimental flags |
| Account & Cloud | Sign-in to your Harmony account, Cloud hours and links |
| Workspaces | Launch profiles, per-workspace defaults, project config |
| Notifications | Notification policy and sound |
| Privacy | Install and update lifecycle telemetry, crash reports |
| Appearance | Theme, custom theme editor, density, skin, fonts, transparency |
| Shortcuts | Rebinding |
| Agents / Harmony | YOLO default, A2A mode, provider probe |
| Tools | Agent tool configuration, including the spawn tools (off by default) |
| Claude integration | Hook bridge install/remove/status, plugin-conflict warning |
| First-run setup | Re-run the onboarding wizard |
| Read Aloud | Text-to-speech engine, voice, model cache, corrupt-model recovery. Preview |
| Dictation | Whisper speech-to-text model, language, push-to-talk timings. Preview |
| LanLink | Pairing, peers, status. Experimental |
| About | Version and build information |
Appearance
Theme changes preview live with a Keep/Revert banner, so you can see a theme applied to your real content before committing to it.
Built-in themes: nala-stream (the default), nala-classic, catppuccin-mocha,
monochrome, stars-and-stripes, red-dynasty, nightowl, void, hinomaru,
taegeuk, gatsby, and two glass themes, azure-glass and magenta-glass. There
is also a custom theme editor with ten tokens and an xterm palette.
Density presets: airy, balanced, dense. If the UI feels cramped or wasteful, this is the setting you want rather than the font size.
Provider probe
Under Agents / Harmony. Reports installed versions, detected authentication state where visible, and the resolved executable path.
IMPORTANT
The probe answers "can I start this program," not "will it produce a turn." Harmony does not hold your provider credentials and cannot report your plan or quota. See account and authentication.
Speech-to-text defaults
| Setting | Default |
|---|---|
| Model | tiny.en |
| Language | en |
| Push-to-talk hold | 250 ms |
| Minimum clip | 0.18 s |
| Segment rollover | 45 s |
Dictation requires Python 3.10+ on PATH. Models download on first use and run
on-device thereafter.
Project configuration
Under Workspaces. A project may carry nala.json (preferred) or the legacy
wmux.json, declaring layout, startup processes with a restart policy, and
supervision burst limits.
SECURITY
Project config is trust-gated, and that gate exists for a reason: the file is content written by whoever wrote the repository. Opening a folder must not be enough to make it start processes.
Settings are typed, not ad hoc
Settings are backed by a typed definition registry with resolve, update, reset, export, import, and migration operations — not loose keys. Exported settings can be reviewed before import, and a preview step exists for that reason.