Configuration reference

Core plugin configuration

These values configure the plugin for the whole DSH profile. Session settings belong to one story instead.

Field	Default	Purpose
rootPath	~/dsh-roleplay	Folder for worlds, avatars, sessions and media
sessionStorePath	~/.dsh/sessions	DSH session store used for the chats attached to roleplay sessions
authoringTools	none	Tools from other plugins that each authoring assistant may use. Groups are worldAuthoring, presetAuthoring, lorebookAuthoring and worldAuthor. An unlisted tool is unavailable. The session doctor never receives these tools
catalogues	none	World catalogues to browse. Each entry has name, API baseUrl, and optionally an apiKeyRef for DSH's credential store. The deprecated apiKey field is used only when the store has no key, then removed on the next save
sync.remote	empty	rclone remote and folder, such as mydrive:dsh-roleplay; empty disables sync
sync.rclone	empty	rclone executable; empty finds rclone on PATH
sync.configFile	empty	Absolute path to the rclone configuration file, which must be outside rootPath
sync.passwordRef	empty	DSH credential-store reference for an encrypted rclone configuration; empty uses ROLEPLAY_SYNC_RCLONE_CONFIG_PASS
sync.intervalMinutes	0	Automatic-sync interval; 0 permits manual runs only
catalogueMaxBytes	256 MiB	Maximum downloaded-world size, checked both compressed and unpacked
catalogueTrashDays	30	Days to keep files replaced by a catalogue update in .trash/worlds/; 0 keeps them indefinitely
importMaxUploadBytes	100 MiB	Maximum character-card upload size
importConcurrency	4	Import and authoring helper calls that may run at once
importReasoning	empty	Reasoning level for import and authoring helpers and authoring chats; empty uses childReasoning, then the chat setting
childReasoning	empty	Fallback reasoning level for roleplay helpers; empty follows the game master
importModel	empty	Model for import and authoring helpers as provider/model; empty uses the chat model
importStructuredOutput	schema	How import and authoring helpers return structured results. schema asks a compatible endpoint to constrain a JSON reply with response_format; LocalAI, vLLM and OpenAI support this form. If an endpoint rejects the schema, that model uses a tool call from then on. tool uses tool calls for every model
importMaxTokens	0	Ceiling on output tokens for one import or authoring-helper request; 0 is a tenth of the model's context window (at least 16000); each job may use a smaller limit
importVariablesPerRequest	25	How many tracked values one import request is asked to declare; a card that tracks more is asked again for the rest
importTemperature	0.2	Import and authoring-helper temperature
importTimeoutMinutes	15	Minutes before an import or authoring helper is stopped without a result; 0 disables the timeout
importDecisions	llm	decision lets the configured decision service preclassify card pieces it can settle confidently; llm sends every piece to the language model
importConfidence	0.7	Confidence required to accept a decision-model classification
authoringApproval	ask	ask requests approval for every world-authoring-chat change; auto writes immediately while retaining replaced or deleted files in the world's .trash folder

All fields except the restart-time sessionStorePath are editable under Settings, Role Play and take effect without restarting. See Sync between devices before pointing more than one installation at the same remote.

State plugin configuration

The state package can be installed on its own. These values configure its database and model-facing tools.

Field	Default	Purpose
journalMode	delete	SQLite journal mode. delete keeps a session easy to copy as one file; wal permits readers during a write
writeTools	false	Register model tools that change records. Reading tools are always registered
diceTools	false	Register dice tools for every roleplay agent. The full bundle instead grants them by agent kind

Session defaults

These defaults are copied into a new story's settings.json. A world or preset may suggest different starting values, and the player can change them for that story.

Turn, memory and context

Setting	Default	Meaning
maxRounds / maxCharacters	3 / 4	Character rounds per player message, and characters activated per round
memory.flaky	true	Apply imperfect perception and story-time fading
memory.recall / memory.lookupCalls	12 / 5	Memories supplied per activation, and maximum lookups in one character decision
memory.lookup / memory.verify	true / true	Let characters query memory, and have the keeper read edits back
memory.backlogRounds	6	Unrecorded rounds that force the keeper to catch up
memory.overRetrieve / memory.associate	4 / 2	Candidate multiplier before selection, and related memories allowed with a strong match
memory.diversity	0.3	Penalty, from 0 to 1, for recalling near-duplicates together
memory.foldedFade	0.5	Retention multiplier after a belief incorporates a perception; 1 leaves it unchanged
context.limitTokens / context.limitRatio	0 / 0.6	Explicit folding limit, or share of the model context used to derive it
context.workingTokens / context.retainRatio	0 / 0.5	Optional cap on the derived limit, and its verbatim share
context.shortTermTokens / context.longTermTokens	1500 / 2500	Short-term and long-term summary targets

World activity and time

Setting	Default	Meaning
authoring.enabled	true	Let the game master ask the world author for missing entities
authoring.perTurn / authoring.thingsPerTurn	2 / 4	Authored characters and places, and ordinary things the GM may introduce, per turn
authoring.consolidateAfter	3	Pending authored entities that trigger consolidation
clock.mode / clock.timeZone	simulated / empty	Story-time mode (simulated or real) and its IANA time zone. At session start, an empty zone uses the host zone for real time, or the world's zone and then UTC for simulated time
clock.roundSeconds / clock.maxSkipDays	30 / 30	Default simulated round duration and maximum days in one clock advance
background.concurrent	false	Let background work overlap the next turn instead of completing first
offstage.perTurn / offstage.perCharacter	2 / 1	Due off-stage characters per turn, and actions allowed for each
offstage.joint / offstage.summary	true / true	Process overlapping situations together, and produce the Meanwhile summary
factions.perTurn / factions.everyDays	1 / 3	Due factions advanced per turn, and their default story-time interval
triggers.enabled / triggers.maxFires	true / 3	Whether triggers run, and how many may fire per turn
triggers.beatTurns / triggers.cueTurns	3 / 2	Turns a game-master beat or character cue waits to be used
triggers.judgedPerTurn	2	Trigger conditions that may require model judgement per turn

Reflection and decision checks

Setting	Default	Meaning
reflection.auto / reflection.everyTurns	true / 20	Reflect automatically after this many turns; 0 disables the schedule
reflection.onSceneChange / reflection.sceneDays	true / 1	Also reflect after a place change or this many story days; 0 disables the time condition
reflection.minTurns / reflection.perRun	5 / 3	Minimum gap between passes and characters processed per automatic pass
reflection.minNew	4	New perceptions needed for an automatic pass; a manual pass needs one
decisions.triggers / decisions.reflection	hybrid / llm	Judge trigger statements and belief evidence with llm, decision, hybrid or shadow. hybrid accepts confident decision-model answers and sends the rest to the language model. Without a configured decision model, it quietly uses the language model alone
decisions.agency / decisions.refusal	off / off	Optional output checks, using the same four modes plus off
decisions.rules / decisions.secrets	off / off	Decision-service-only checks of allowed actions against rules, and outputs against hidden lore
decisions.gateThreshold	0.5	Probability at which a decision answer counts as a fault; the 0.15 below it is uncertain
decisions.confidence	0.7	Confidence required before a decision answer is accepted instead of handed to the language model

The modes and fallback rules are explained in Optional capability integration.

Models, prompts and sampling

Setting	Default	Meaning
models.characters, models.records, models.suggestions, models.authoring	empty	Model for that helper as provider/model; empty uses the chat model
models.background	empty	Background-character model; empty falls back to models.characters
sampling.master / sampling.characters	0.7 / 0.8	Game-master and character temperatures
sampling.suggestions / sampling.authoring / sampling.records	0.2 / 0.4 / 0.1	Temperatures for suggestions, authoring and records
sampling.topP / sampling.topK / sampling.minP	1 / 0 / 0	Additional sampling controls; defaults are not sent
sampling.frequencyPenalty / sampling.presencePenalty / sampling.repetitionPenalty	0 / 0 / 1	Penalty controls; defaults are not sent
sampling.tools	required	Tool-choice mode for helpers that must submit a result
sampling.masterMaxTokens	12000	Maximum output tokens for one game-master request, including reasoning
sampling.maxTokens / sampling.shortMaxTokens	8000 / 2000	Output ceilings for general and short helper requests
sampling.headroom	2	Multiplier applied to a helper's estimated answer size before its ceiling
prompts.language	en	Shipped prompt language: en or zh
reasoning.children	empty	Helper-agent reasoning level; empty uses the plugin fallback

Interface, rules and media

Setting	Default	Meaning
interface.pack	empty	Empty uses the world's first pack, none disables packs, and a name selects one
rewind.keepSnapshots	30	Recent state snapshots retained for taking the story back
rules.own	empty	Additional rules supplied by the player for this story
preview.enabled	true	Show an early scene preview while the full turn runs
suggestions.count	3	Suggested next actions shown after a turn
dice.visibility	card	card shows rolls in a collapsed card; hidden shows no roll card
tools	none chosen	Optional tools by agent kind (gameMaster, characters, records) and name; unknown names are retained and ignored
images.mode	manual	off, manual or every-turn
images.references	prefer-reference	prefer-reference, require-reference or text-only
images.reuse / images.highRes	true / false	Reuse existing pictures, and optionally render a larger second result
images.interpret	true	Describe uploaded pictures for the story
images.interpreterModel / images.captioner	empty / empty	Models for uploaded-image interpretation and generated-image captions; empty uses the chat model
diagnostics.keepRuns	false	Keep the latest 60 helper traces, including private character context, in the session folder

The UI shows current numeric bounds. Unknown fields are rejected. Provider-specific embedding request options are stored under embedding.options.