Settings

One screen for every app's knobs. Each toryo app that has tunable settings describes them itself, naming what fields it has, their types, their ranges, and their defaults, and board renders that description as a form.

The consequence: board knows nothing about any particular app's settings. An app that gains a new setting gains it in board with no change to board.

What you can tune#

• Dispatch: how many agent jobs run at once. Raising it applies immediately, since the queue re-reads the cap every tick rather than at startup. You can also set whether it admits new work at all (Pause admission, see Dispatch), and how many held conversational sessions stay open (Max held sessions, default 3).
• Sequence: how many times a repeatedly-failing step retries before escalating to you, how API errors back off, and the directory a run that names no project is rooted in. There is deliberately no fallback directory for that last one, so a project-less run always has a location you chose.
• Sequence, board defaults: which run statuses and which project the Sequence screen opens filtered to, so you start on the runs you actually work from rather than every run in every project.
• Worker: which agent harnesses this machine offers at all, which one is the default, and the context-optimization flags that govern how a session manages its own context, covering reading a file once, caching command output, guarding against verbose output, diffing re-reads, and compressing tool results. The harness switches have a rule of their own: at least one must stay enabled, and you cannot disable the one your default names. You can also set how many subagents one Claude Code session may run at once (maxSubagents, default 3, 0 for no limit), counting the agents a Claude Workflow starts. See Harnesses.
• Usage: how often to poll each harness for its rate limits, which ones to show, and when a meter turns red. See Usage.
• Janitor: whether the scheduled sweep runs, whether it deletes or only reports (dry run), and how long each kind of leftover is kept. How often the sweep runs is set by the system maintenance lane, not here; the old interval setting is gone.
• Messaging: how long an unanswered message lives, and how long a repeating notification stays quiet after its first ping (notifyWindowSec, default one hour, 0 to ping every time). See Messaging.

Foreman, scheduler, and project have no settings yet; their daemon intervals and ports are start-up flags today.

Validation happens in both places#

The same validator runs in board and in the CLI, so a value board accepts is a value the app accepts. A setting that references something else, such as the default project for the Sequence screen, is checked against the project registry when written, and an unknown handle is refused with the valid ones named.

Board is forgiving on read, though: a project you have since removed is treated as no filter rather than an error, because removing a project hides it from every read, and a stale reference is routine.

Usage gauges#

board also polls each harness for its rate limits, showing how close you are, so you can see before dispatching a large batch whether you have the headroom for it. The full breakdown has its own screen and its own page: Usage.

In board#

The Settings screen groups every app that exposes settings. Each field carries its own help text from the app's own description of it.