YOUR NEXT STEP

Buddy Chat

Instructions for the upcoming NeuralCompanion interface.

Buddy Chat lets one or more buddy Personas participate naturally in the normal Main Chat. It is a conversation layer, not a separate room. Replies still use the main chat session, and the selected buddies can share the reply or speak in turn depending on the participation settings.

Buddy Chat Is Not A Room

Use Buddy Chat when you want several distinct voices and personalities inside the ordinary conversation. Use Local Room for a separate moderated group that has its own transcript and turn-taking. Use Multi Persona Roleplay for an isolated story feed with a narrator and Start Play.

Setup Versus Activation

Open Addons → Buddy Chat. The workspace has Setup, Overview, Buddies, Voices, Avatars, Providers, and Advanced sections.

  1. Addon setup: enable Buddy Chat in the Addons workspace and restart if the toggle says Enabled after restart.
  2. Group setup: build or load a group in the addon’s Setup tab. A staged preview is not saved until Apply group.
  3. Runtime activation: open Overview, turn on Enable Buddy Chat, and select Save. This is separate from the Addon Manager’s next-launch toggle.

The status chip in the addon workspace can say Loaded this session while Buddy Chat itself is still disabled in Overview.

Build A Group

Open Setup and either:

  • choose a complete quick start such as Reliable Guide, Supportive Duo, Expressive Circle, or Magnetic Company, or
  • use Build my group to choose the buddy count, participation style, and personality mix, then select Preview guided group.

Review the staged names, roles, descriptions, prompts, speaking styles, and participation values. Apply group commits the staged setup; Cancel preview leaves the current saved group unchanged. Undo last setup is available after an applied setup.

Refine a staged group

Create a preview first, then use Optional LLM refinement in Setup:

  1. Choose Refine scope: Entire group, One buddy, or One field.
  2. For the last two scopes, select Refine persona. For One field, also select Refine field; that selector does not limit One buddy.
  3. Enter a concrete Refinement instruction, such as “Make this buddy’s speaking style shorter and more curious.”
  4. Select Refine with LLM and read Refinement result. Use Cancel refinement to stop an unwanted request.
  5. Inspect the revised staged fields and changes before Apply group.

Refinement uses the configured LLM route. It edits the preview rather than silently applying a saved group. It preserves buddy IDs and does not configure providers, credentials, voices, or avatars. If a model returns an invalid or out-of-scope response, read the result and retry with a narrower instruction; the failed response is not a new saved group.

Edit Buddies

Open Buddies to review each buddy’s active switch, name, role, description, speaking style, and system prompt. A deactivated buddy keeps its configuration for later use. Use Save in Overview after deliberate edits.

Keep roles distinct enough to be useful. For example, one buddy can explain a single idea at a time while another asks short follow-up questions. Adult/NSFW mode, when present, is opt-in and requires confirmation; it is not required for normal Buddy Chat.

Voices

Open Voices to enable Use voice sample for a buddy and choose a local audio reference. Voice routing depends on:

  • a desktop TTS backend that supports reference audio,
  • a saved, valid reference file,
  • and recognized buddy or speaker segments in the reply.

Assigning different samples does not by itself prove the route. Test a short exchange and confirm that the intended buddy voice is actually used. Typing a buddy name in a prompt is not a routing test.

For creating reusable references, use Voice Studio. Voice Studio saves a WAV to its library; you still assign it in the Persona or buddy voice settings.

Language Providers

Overview → LLM provider mode offers:

  • Use Main LLM Runtime — the recommended, simplest first test. Buddies use the active Main Chat provider/model.
  • Use Buddy Provider — a shared provider configured in Providers.
  • Use Per-Persona Providers — individual provider/model settings per buddy.

Provider overrides need valid models, endpoints, and credentials. Use Refresh shared models or Test Buddy provider where offered, then verify one ordinary reply before adding more complexity.

Activate And Tune

In Overview:

  • Enable Buddy Chat turns the layer on for normal chat.
  • Reply mode defaults to Natural main conversation - recommended. Direct Buddy-generated replies - advanced changes how replies are produced.
  • Max buddies per reply limits how many buddies can contribute.
  • Force buddy every N replies can keep a quieter cast involved.
  • Show active persona window is optional; the active Persona is already represented in the Conversation workspace.

Select Save after changes. Then send a normal message such as:

Teacher, explain how a rainbow forms; beginner, ask one useful question.

Inspect who spoke, whether the reply still fits the main conversation, and whether the response was generated once rather than as several independent turns.

Avatars

Avatars contains per-buddy image and generation controls. Avatar generation requires a configured Visual Reply provider and may consume provider credits. Use a plain text reply first; enable avatar generation only after the group and voice routing are working.

Prompt Layers And Conflicts

Setup → Shared prompt layers holds the shared Buddy Chat override and the normal and Adult/NSFW conversation prompts. Advanced holds Override main persona instructions while Buddy Chat is enabled, Use recommended override, and the editable Buddy system override prompt. Use the override when the single active Persona’s instructions are suppressing the buddy cast. Review the shared layers if the buddies sound too similar or one buddy dominates.

These settings change prompts, not provider behavior or the main chat history. Keep a working configuration saved before experimenting.

Troubleshooting

Symptom Check
Buddy Chat tab is missing Enable the addon in Addons and restart NC if requested
Group setup disappears A staged preview is only saved by Apply group
No buddies speak Enable Buddy Chat and Save; check each buddy’s active switch, Max buddies per reply, and Force buddy every N replies
One Persona dominates Review Override main persona instructions, Buddy system override, shared prompt layers, and buddy participation values
Voices do not change Confirm a reference-capable TTS backend, valid samples, and recognized buddy segments; test a short reply
Refinement does not change the saved group Build a staged preview, choose the scope and buddy/field, read Refinement result, review the proposed changes, then Apply group
Provider override fails Validate the shared or per-persona provider, model ID, endpoint, and credentials in Providers

Current Limits

  • Buddy Chat uses the normal Main Chat runtime, not a separate room or memory store.
  • It does not load multiple TTS backends. Per-buddy voices require the active backend to support reference audio.
  • Provider overrides can use different models, but they do not create a separate parallel conversation for each buddy.
  • Avatar generation depends on Visual Reply configuration and provider cost.
  • The quality of a generated group depends on the model’s ability to follow speaker and participation instructions; it is not deterministic.