Build a character that can reason about its world.

Start here

Quick start

1. 01
   Launch AppStarter.

   Choose a NoruCore workspace and review the local-use agreement.

2. 02
   Let AppStarter verify the installation.

   It selects packages for the detected operating system and architecture, downloads missing files and opens NoruCore.

3. 03
   Select a local model.

   Open Settings from Home, choose a detected model or select a folder containing .gguf files.

4. 04
   Create a Book.

   Add a Room description, structured objects and the Persona values you care about.

5. 05
   Open Chat.

   Talk naturally, request actions, inspect different chat sessions or return to Book configuration at any time.

Installation and launch

AppStarter

AppStarter is the entry point to NoruCore and an integral part of the application. Every launch starts there: it prepares the workspace, verifies the installed files, applies available updates and then opens the main NoruCore interface.

Your library

Home

Home shows every saved Book in the active workspace. A Book contains one configured scenario, its Room, Persona and independent chat sessions.

• Add BookCreate a new empty Book and open its configuration.
• SearchFilter the visible Book cards by name.
• OpenClick a configured Book card to enter its latest Chat.
• ManageRight-click a Book card to Rename it, open Book configuration or Delete it.

Book names must be unique inside the workspace. Returning from Book configuration saves the current Book before Home is reopened.

Application preferences

Settings

Open Settings from Home. Changes are stored in the shared workspace and remain available when NoruCore is launched again through AppStarter.

Scenario authoring

Book configuration

A Book is the boundary around a scenario. The top navigation switches between Room, Persona and Training. The adjacent button opens Extra in Room or Persona, and becomes the Training objective selector in Training.

Use Back when you finish. NoruCore validates the Book name, rebuilds the Room grounding registry and saves the Book before returning Home.

What exists in the scene

Room

The Room combines a free description with structured object cards. The description gives atmosphere and broad context; cards give the Brain dependable information about what an object can do.

Create an object

1. Type an object name in the search field and select Add object.
2. Choose its functional type from the custom menu.
3. Set Interactable and Movable.
4. Right-click the card when you need to delete it.

Object cards can be dragged into the Room description. Dropping one inserts a stable token such as {{wooden chair}}, keeping the text connected to the authored object.

Who the character is

Persona

Decision influences

Mind and relationship

These controls influence Brain decisions. Use fewer, deliberate values when you want a character that is easier to understand and tune.

• Base emotionsJoy, Sadness, Anger, Disgust, Fear and Surprise establish the character's starting emotional balance.
• TraitsConscientious, Prudent, Defiant, Impulsive, Socially conditioned and Empathetic influence action costs and behavior.
• Relationship with youChoose Stranger, Acquaintance, Friend, Ally, Partner, Family, Rival or Enemy, then refine Trust, Affection, Respect and Familiarity.
• PrinciplesAdd a canonical principle such as Tell the truth, Respect boundaries or Keep promises, then set its strength. Right-click to delete it.
• Character ValuesAdd values such as Compassion, Honesty, Freedom, Justice or Security. A positive setting makes the character more likely to support that value; a negative setting makes the character more likely to oppose it. Values closer to either end have a stronger influence, while the center remains neutral.

Repeatable behavior

Habits

Select Add to create a Habit card, then click the card to configure one action and one known target. Available actions come from the Persona's real action catalog; targets are limited to known people, the person interacting with the character, Room objects and inventory items that fit the selected action.

Time fields use the 24-hour HH:MM format: for example, enter 07:30 for half past seven in the morning or 19:45 for quarter to eight in the evening. Intervals accept minutes, hours or days, and specific days can be selected individually. Right-click a Habit card to delete it.

Goals with conditions

Quests

A Quest stores one or more structured objectives. Select an action and a compatible known target, then press Add objective to keep that pair and prepare another objective.

• ImportanceLow, Normal, High or Critical influences priority.
• RepeatableAllows the Quest to become immediately repeatable after completion.
• Multiple objectivesEvery saved objective must succeed for the Quest to complete.
• PrerequisitesCompleted Quest, owned item, current Room, readable world state or none. Multiple prerequisites use AND logic.

Prerequisite options are drawn from known Book data. Internal runtime identifiers are translated into readable labels in the menu. Click a prerequisite's value and choose Delete to remove it; right-click the Quest card to delete the whole Quest.

Carried and worn items

Equipment and inventory

The character contains seven fixed cards: Head, Body, Legs, Feet, Hands, Accessory and Equipment. Click a card to edit one slot.

• Each slot can contain multiple named items.
• Drag an item inside the popup list to change its saved position. This order is preserved and determines how items are presented and consulted within that slot or the general inventory.
• The Equipment card represents general inventory; the other cards represent worn slots.
• Chat actions can inspect, equip, unequip, use, give or drop items when the current state allows them.

Book presentation

Extra

Extra controls the background and sound associated with the current Book. It uses only assets included with NoruCore; it does not import arbitrary media.

Guided correction

Guided Training

Press and hold Persona to open its context menu, then select Training. The former Extra button becomes a selector for five focused modes. When you change mode, the visible Training conversation is cleared from the interface so the next session has one explicit objective. Corrections already applied to the character remain saved in their dedicated profiles.

Feedback flow

• LikeConfirms the generated result for the active objective. Behavior Like confirms the trace without changing Traits.
• DislikeOpens a correction panel below that response. Adjust the controller or formatting preset, then Apply & Retry.
• Edit responseRight-click a Voice, Delivery or Formatting response to rewrite a selection or the complete response, then apply or retry.
• ExpressionsSelect text and right-click to mark it as preferred or avoided. Explicit selection is required.

Actions Training

The experimental Actions mode shows the action interpreted by the Brain, why it was considered and its real alternatives. Select Custom to correct Understanding, Decision or both through two closed menus with four fixed choices each. The corrections are saved separately in the Persona.

Actions uses the active Room, accepts unique partial object names such as sofa for Purple Sofa, and follows the available GOAP action catalog. Its sandbox never executes the action or changes the world. For now, its corrections are recorded and restored with the Persona, but are not yet used by the Brain when making later decisions.

Role-playing runtime

Chat

Chat combines the configured and trained Persona, current Room, grounded objects, inventory, Habits, Quests, relationship and learned state. The Analyzer prepares your message for the Brain, which decides how the character should behave based on the configuration and Training you have applied. The selected local model then writes the final response around that decision.

• Natural conversationAsk questions, describe events or request supported actions in English.
• Grounded actionsThe Brain checks target identity, object affordances, accessibility, mobility, inventory and world state before committing actions.
• Multiple chatsOpen the Chat selector to create a new Chat, switch between saved Chats or delete the current one.
• Turn controlsEach history entry offers Copy, Delete turn and Rewind to here.

Chat action feedback

After one successful action, send a separate English reaction such as Great!, Awesome!, Bad! or Yuck! to make that action more or less preferred in similar future decisions. In this experimental version, the Brain recognizes only a fixed built-in vocabulary: these phrases cannot be added, removed or edited from the application. Matching ignores capitalization and punctuation, but the reaction must correspond to one of the supported phrases and cannot contain a new instruction. Each learned preference remains linked to its feedback message; deleting that message removes only the associated learning and preserves feedback from other messages.

Continuity without an endless prompt

Memory and context

NoruCore stores structured character and gameplay information separately from the raw model prompt. This allows important learned facts, relationship state, inventory and Book configuration to persist without sending the entire history on every turn.

The prompt budget reserves output space and progressively compacts older conversation when necessary. If the configured model still cannot fit the required context, NoruCore blocks the request and asks you to start a new Chat or continue with less recent context.

Training conversations are temporary working sessions and are not added to normal Chat history. NoruCore saves only the changes you explicitly confirm: Voice, Delivery and Formatting update their dedicated communication settings, while Behavior corrections update the character's six core Traits. The complete Training conversation is cleared when its session is reset.

Honest Preview boundaries

Current limits

• No geometric world simulation.The Brain understands authored objects and functional relationships, not exact coordinates or visual geometry.
• No automatic room topology.A door can be authored as a passage, but the current UI does not define exactly which Room it leads to or where every object sits relative to another.
• English is the supported analysis language.Other languages may be narrated by some models, but action understanding is not guaranteed.
• Not every sentence becomes an action.If a request is ambiguous, outside the supported action vocabulary or missing a clear target, NoruCore may ask for clarification, continue it as conversation or reject it without changing state.
• Dialogue and state can diverge.The language model narrates a Brain decision but cannot directly change authoritative state. In this alpha Preview, its wording may occasionally imply an action that the Room, inventory or other systems did not accept.
• Complex references remain experimental.Negation, several actors in one sentence, indirect references and long chains of dependent actions can still be interpreted incorrectly. Prefer explicit names and shorter requests when accuracy matters.
• Model quality matters.The Brain constrains decisions, but a very small or unsuitable narrator model may still produce awkward wording or imperfect prose.
• Local performance varies.Loading and response time depend on model size, context length, available RAM/VRAM and local inference support.
• Preview stability.The application is still experimental. Save important Books and keep AppStarter-managed backups when testing new releases.

Common setup problems

Troubleshooting