Dialectic

Dwemer Dynamics HomeOverviewInstallationConfigurationRoleplay SettingsFallout SettingsLarge Language ModelsLocal LLMsText-to-SpeechSpeech-to-TextModders GuideFAQRemote Hosting GuideLinux Setup GuideLog Files

⬇️DownloadDiscordDiscord🎥Video GuideYouTubeYouTubePatreonPatreon💻Server Code💻Mod Code

Dialectic Large Language Model Services

Language model connectors generate dialogue and support summaries, profiles, memories, relationships, narration, scene classification, and Director Mode.

Services

Open Configuration - LLM to create, test, import, export, and assign connectors.

OpenRouter

OpenRouter is the recommended first setup. One API key provides access to many hosted models and works with Dialectic's Standard, Fast, Powerful, and Experimental presets.

Create an account at OpenRouter, add credit, create an API key, and enter it in Quickstart or API Keys.

Player2

Player2 provides a local application-managed connector. When the Quickstart Player2 option is enabled, all four in-game model presets use the model selected in the Player2 application.

OpenAI

The OpenAI connector uses the OpenAI-compatible chat-completions format. Store the API key in API Keys, then choose a compatible model and test the connector before assigning it.

Google

The Google connector uses Google's OpenAI-compatible endpoint. Model access and pricing are controlled by the linked Google account.

Groq

Groq provides fast hosted inference for supported models through an OpenAI-compatible API.

Nano-GPT

Nano-GPT is available as another hosted OpenAI-compatible service and can be configured from the normal connector editor.

Local Models

Run LM Studio inside DwemerDistro using the launcher's LLM Studio component. The Local LLM Guide walks through downloading, loading, testing, and connecting a model. It also covers separate Windows LM Studio and KoboldCPP servers.

The bundled engine uses http://127.0.0.1:1234/v1/chat/completions from DialecticServer in the same DwemerDistro. A separate Windows app needs an address DialecticServer can reach.

Guided local-model setup is available in Quickstart. Review its target and scope before applying a model; for a single test profile, follow the guide's manual connector steps.

Use Your Own LLM Provider

OpenRouter is optional. Dialectic can connect to other hosted or local services through a Custom connector.

This is the advanced route, for full manual control of the endpoint, driver, and model. For a model on your own PC, try Local Models first.

Before You Start

Get these details from your provider:

  • An API key, unless the service runs locally without one.
  • The complete chat-completions URL.
  • The exact model name the provider expects.

Create the Connector

  1. Open API Keys. Save the provider's key in a preset entry, or add a named custom key.
  2. Open Configuration - LLM, select New Connector, then choose Custom.
  3. Enter a clear Name, the complete URL, and the provider's exact Model name.
  4. Choose the matching Driver. Common choices include OpenAI JSON for OpenAI-compatible services, Google OpenAI JSON for Gemini, and Groq JSON for Groq.
  5. Select the API key you saved.
  6. Leave Reasoning Model Fix, Enforce JSON, JSON Schema, Prefill JSON, and Disable Streaming unchanged unless the provider specifically requires them.
  7. Click Save, then Test. Do not assign the connector until the test succeeds.

Example: Google Gemini

Gemini is one example of a provider you can add manually:

  • API key: Save your Google AI Studio key under the Google preset or a named custom key.
  • URL: https://generativelanguage.googleapis.com/v1beta/openai/chat/completions
  • Driver: Google OpenAI JSON
  • Model: Use the exact current model name shown in Google AI Studio.

The built-in Google connector is simpler for most users. Use Custom when you want manual control over the endpoint and model. See Google's OpenAI compatibility guide for its current endpoint and model guidance.

Use It in Dialectic

Open Profiles and select the connector for Standard, Fast, Powerful, Experimental, or another profile role. For shared work such as summaries, memories, dynamic profiles, scene classification, Director Mode, relationships, or world knowledge, assign it under Global Settings.

If the Test Fails

  • 401 or 403: Check the selected API key and the provider account.
  • 404: Check the full URL and model name.
  • No usable response: Make sure the selected driver matches the provider's response format.
  • Invalid JSON: Try a model that handles structured responses well. Only enable the advanced JSON options when the provider supports them.

In-Game Model Presets

Standard, Fast, Powerful, and Experimental are server-owned model selections. The Toggle LLM Model hotkey asks DialecticServer for the available options and changes the active mode without storing a second conflicting value in the plugin.

Global Connector Roles

Separate connectors can handle the player, summaries, middle-term memory, dynamic profiles, scene classification, Director Mode, relationships, and custom world knowledge. Smaller fast models are often enough for background tasks.

Test each connector from its editor before assigning it globally or to a profile. A working chat model still needs to follow Dialectic's structured JSON response schema for actions and multi-line dialogue.