Settings

The dashboard is configured via a TOML file at ~/.agent-dashboard/settings.toml (or $AGENT_DASHBOARD_DIR/settings.toml if overridden). From a repo checkout, the installer copies this from settings.example.toml when the destination file does not already exist. Any missing keys fall back to sensible defaults — you only need to include the settings you want to change.

Full reference

[banner]
show_mascot   = true   # show the axolotl pixel art
show_quote    = true   # show the daily quote

[notifications]
enabled       = false  # enable desktop notifications from adapter hooks
sound         = false  # play alert sound on attention events
silent_events = false  # show notification for non-alerting stops

[debug]
key_log       = false  # write key/mouse/focus events to debug-keys.log

[experimental]
ascii_pet     = false  # show animated ASCII pet in the left panel
dino_game     = false  # enable Chrome-style dino runner game (Shift+G)

[usage]
rate_limit_poll_seconds = 60  # how often to fetch rate limits from Anthropic API (0 = disable)

[effort]
plan    = "high"  # thinking-effort level pinned while permission_mode='plan'
default = "high"  # thinking-effort level pinned at spawn and restored on plan exit

[harness]
default = "claude"  # active coding-agent harness: "claude" or "codex"

[harness.codex]
model = ""                    # e.g. "gpt-5.5" — passed as codex --model
approval = ""                 # e.g. "on-request" — passed as codex -a
sandbox = ""                  # e.g. "workspace-write" — passed as codex -s
default_reasoning_effort = "" # e.g. "high" — passed as -c model_reasoning_effort=high

The [effort] levels feed the /effort slash command Claude Code accepts (minimal | low | medium | high | max). The agent-state-fast adapter hook swaps in plan when the agent enters plan mode (EnterPlanMode) and restores default on exit. The feature, fix, and refactor skills additionally declare effort: max in their frontmatter, which Claude Code pins for the skill’s lifetime when the skill is invoked as a slash command inside an existing session. For Codex sessions, max is clamped to high (the top of Codex’s reasoning-effort scale).

The [harness] section selects which coding-agent binary backs newly-spawned sessions. "claude" uses Claude Code (default; reads ~/.claude). "codex" uses Codex CLI (reads ~/.codex) and applies [harness.codex] model, approval, sandbox, and reasoning-effort settings. These are defaults — the New Agent flow lets you override harness, model, and thinking effort per spawn.

The New Agent model and effort pickers hint what their (default) options will use. For Codex, [harness.codex].model and [harness.codex].default_reasoning_effort win over ~/.codex/config.toml because they are the effective dashboard spawn defaults. For Claude effort, [effort].default wins over ~/.claude/settings.json. Native config reads are cached for 1 hour and can be refreshed with Ctrl+R in the TUI picker or the web refresh button.

Settings table

Section	Key	Default	Description
`banner`	`show_mascot`	`true`	Show the axolotl pixel art in the banner
`banner`	`show_quote`	`true`	Show the daily quote in the banner
`notifications`	`enabled`	`false`	Enable desktop notifications from adapter hooks
`notifications`	`sound`	`false`	Play alert sound on attention events
`notifications`	`silent_events`	`false`	Show notification for non-alerting stops
`debug`	`key_log`	`false`	Write key/mouse/focus events to `debug-keys.log`
`experimental`	`ascii_pet`	`false`	Show animated ASCII pet in the left panel
`experimental`	`dino_game`	`false`	Show Chrome-style dino runner game in the left panel (Shift+G to toggle)
`usage`	`rate_limit_poll_seconds`	`60`	How often (in seconds) to fetch rate-limit data from the Anthropic OAuth API. Set to `0` to disable.
`effort`	`plan`	`"high"`	Thinking-effort level pinned while the agent is in plan mode. One of `minimal`, `low`, `medium`, `high`, `max`.
`effort`	`default`	`"high"`	Thinking-effort level pinned at spawn and restored when the agent exits plan mode. Same value set as `plan`. When set, this is the Claude New Agent effort hint.
`harness`	`default`	`"claude"`	Active coding-agent harness. `"claude"` runs Claude Code; `"codex"` runs Codex CLI.
`harness.codex`	`model`	`""`	Model passed to `codex --model`. Leave empty to inherit Codex’s default. When set, this is the `(default)` hint shown in New Agent. Example: `"gpt-5.5"`.
`harness.codex`	`approval`	`""`	Approval policy passed to `codex -a`.
`harness.codex`	`sandbox`	`""`	Sandbox mode passed to `codex -s`.
`harness.codex`	`default_reasoning_effort`	`""`	Reasoning effort passed as `-c model_reasoning_effort=<level>` for opted-in skills. One of `minimal`, `low`, `medium`, `high` (`max` is clamped to `high`). When set, this is the Codex New Agent effort hint.

Environment variables

Variable	Description	Required
`AGENT_DASHBOARD_DIR`	Override default state directory (`~/.agent-dashboard`)	No
`EDITOR`	Editor command for opening agent directories (default: `code`)	No
`API_NINJAS_KEY`	API key for quote-of-the-day	No (falls back to built-in quotes)
`GOOGLE_CLIENT_ID`	Google OAuth client ID for mobile companion authentication	No
`GOOGLE_CLIENT_SECRET`	Google OAuth client secret	No
`GOOGLE_ALLOWED_EMAIL`	Email address allowed to access the mobile companion	No