Mind configuration

A mind’s configuration lives in home/.config/. Most of it is editable from the dashboard (the mind’s Settings section), but the files are plain JSON and a mind can read and rewrite its own — part of how a mind understands and shapes its own framework.

volute.json

Volute-level configuration: identity, presentation, rhythm, and budget. Read by the daemon.

Field	Type	Purpose
identity	{ privateKey, publicKey }	The mind’s Ed25519 keypair (see Identity)
profile	{ displayName?, description?, avatar? }	Display name, description, and avatar path (relative to home/)
schedules	Schedule[]	Scheduled messages and scripts (see clock)
sleep	SleepConfig	Sleep/wake cycle configuration (see below)
model	string	Model override
thinkingLevel	"off"…"xhigh"	Reasoning depth
spendCap	number	Spend cap in USD per budget period. At 80% the mind gets a heads-up; at 100% messages sent to it are held (not dropped) until the period resets, while its schedules and its own actions keep running
spendCapPeriodMinutes	number	Length of the budget period in minutes (default 1440)
echoText	boolean	Echo the mind’s text output back to it
unescapeNewlines	boolean	Unescape \n in incoming messages

Each entry in schedules has an id, enabled, and either a cron expression or a one-time fireAt timestamp, plus a message (or messages pool) or a script. Optional thread targets a thread, and whileSleeping (skip | queue | trigger-wake) controls behavior during sleep.

sleep

{
  "sleep": {
    "enabled": true,
    "schedule": { "sleep": "0 23 * * *", "wake": "0 8 * * *" },
    "wakeTriggers": {
      "mentions": true,
      "dms": true,
      "channels": ["discord:my-server/urgent-*"],
      "senders": ["alice"]
    }
  }
}

schedule holds sleep and wake cron expressions. wakeTriggers decides what wakes a sleeping mind — mentions and dms both default to true, while channels and senders (glob patterns) are opt-in. See Sleep.

config.json

SDK-level configuration, loaded by the mind’s server at startup.

Field	Type	Purpose
model	string	Model the mind runs on
compaction	{ maxContextTokens? }	Token threshold that triggers compaction
compactionMessage	string	Message shown to the mind before compaction
sessionIdleMinutes	number	Idle minutes before a session’s subprocess is reaped (default 30, 0 disables)
memory.recollection	{ enabled?, coldResetMinutes? }	Claude template. enabled (default true): seed the mind’s consolidated memories ahead of the verbatim tail at every seam; false seeds the tail only. coldResetMinutes (default 55, 0 disables): idle minutes after which a session is re-seeded before its next turn, since past the prompt cache’s lifetime that turn would re-read the whole context anyway. Only a session bigger than its seed resets; with sessionIdleMinutes: 0 a live session never cold-resets (only one re-created after a restart does)
continuity	{ seedTokens? }	Size of the verbatim tail a seeded session carries across a seam (wake, rotation, cold reset). Claude-template seams also carry the mind’s recollection ahead of the tail. Default: claude 10000 when recollection arrived, 30000 when it didn’t; pi and codex 30000. 0 disables seeding
logLevel	"error"…"debug"	Log verbosity
subagents	Record<string, …>	Named subagent definitions
thinking / effort / thinkingLevel / reasoningEffort	—	Template-specific extended-thinking controls passed to the SDK

routes.json

Message routing rules — which channels a mind handles and where their messages go. See Routing for the full format.