Everything you can configure

LLM.json

Turns the AI layer on and points it at your model. Every reaction has a hand-written fallback line, so the plugin runs without this file being touched at all — fill it in when you're ready for live-generated dialogue.

Every field here has a placeholder value out of the box (pointing at nothing reachable) — set your own Url/Model/ApiKey before enabling. Options: a local client like LM Studio, a hosted provider (DeepSeek, Qwen, etc. — set Url, Model and ApiKey), or the Player2 API via UsePlayer2API.

Reaction.json

One on/off switch per reaction category, plus cooldowns for the ones that fire on a timer rather than an event.

Orchestrator.json

Governs the queue that stops Buds from spamming chat — only change these for performance tuning.

Debug.json

Conversation.json

Work.json

Field sizes, work pacing, tree root tuning and mining growth timing — this one is documented in full on the Work Stations page, right next to the mechanics it controls.

Bud Registry (buds/*.yml)

Buds are fully data-driven. Each companion is defined by a YAML file at buds/<id>.yml (id, displayName, color, npcTypeId, weaponId, armorId, pronoun, favoriteDay, promptKey, sounds, plus a workRole and restPosition for Workstation binding). Adding a new Bud only needs a new YAML file plus matching game assets — no plugin rebuild.

buds/roster.yml lists defaultBuds (at most 3) — the set /bud create spawns without an id, and what the Roster card summons. Buds outside the roster are still summonable individually.

Prompt management

LLM system prompts and each Bud's personality/fallback text live as external YAML under the mod's prompts/ folder — edit them to change what a Bud says without touching any code. They're copied from the packaged defaults on first server start.

Content versioning (versions.yml)

versions.yml in the mod's data directory is the single place that tracks whether your server's prompt/Bud content is up to date with the packaged version:

budVersion: 2
promptVersion: 2
excludedPrompts:
  - "buds/gronkh.yml"
excludedBuds:
  - "gronkh.yml"

excludedPrompts/excludedBuds (paths relative to prompts//buds/) mark files to keep untouched during an automatic update — for content you've customized yourself, at your own risk. The plugin only ever updates the version numbers here when syncing content, never the exclusion lists, so your exclusions survive every automatic update. An explicit --reset command still ignores exclusions and overwrites everything.

⚠️ Using reasoning models

If you're pointing LLM.json at a reasoning model (e.g. DeepSeek-R1, Qwen-Reasoning):

• Disable "thinking" — these models generate many <think> tokens that can cause delays or cut-off messages. Prefer a non-thinking model, or disable the thinking phase in your provider's settings.
• Raise the token limit — if messages get cut off, increase MaxTokens to at least 200.
• The plugin does try to filter <think> tags automatically, but a native no-thinking model gives the best experience.