Skip to content

Presets

Change inference parameters, embedding parameters and global system prompt overrides. The Presets group in the settings sidebar holds three pages: Inference, Embeddings and System Prompts.

Inference

Advanced settings. Use with caution.

If these settings don't mean anything to you, you probably shouldn't be changing them. They control the way the AI generates text and can have a big impact on the quality of the output.

This document will NOT explain what each setting does.

App settings - Presets - Inference

If you're familiar with editing inference parameters from other similar applications, be aware that there is a significant difference in how TaleMate handles these settings.

Each preset in a group belongs to one action type. Agents take different actions, and based on that action one of the action types is selected.

That means that ALL action types are relevant and will be used at some point.

For example analysis will use the Analytical action type, which is configured to be less random and more deterministic.

The Conversation action type is used by the conversation agent during dialogue generation.

The other action types are used for various creative tasks.

See Inference Configuration for the parameters and the Action Type selector of the pinned inference panel.

These are all experimental and will probably change / get merged in the future.

Embeddings

App settings - Presets - Embeddings

Allows you to add, remove and manage various embedding models for the memory agent to use via chromadb.

Pre-configured Embeddings

all-MiniLM-L6-v2

The default ChromaDB embedding. Also the default for the Memory agent unless changed.

Fast, but the least accurate.

Alibaba-NLP/Gte-Base-En-V1.5

Sentence transformer model that is decently fast and accurate and will likely become the default for the Memory agent in the future.

OpenAI text-embedding-3-small

OpenAI's current text embedding model. Fast and accurate, but not free.

Requires an OpenAI API key.

OpenRouter openai/text-embedding-3-small

The same model, served through OpenRouter's embeddings endpoint. Use this if you already run Talemate on OpenRouter and do not want a separate OpenAI account.

It uses the OpenRouter API key from Connections → API Keys, so there is no extra key to set here. See the OpenRouter API setup. You do not need an OpenRouter client for this — the key is enough.

The key is required

With this embedding selected and no OpenRouter key configured, the Memory agent reports the missing key and scenes will not load until you set one. This is the same behaviour the OpenAI embedding has without an OpenAI key.

OpenRouter serves other embedding models as well. To use one, add an embedding with the OpenRouter type and enter the model id, for example qwen/qwen3-embedding-8b.

Adding an Embedding

You can add new embeddings by clicking the Add new button.

Select the embedding type first. The form then asks for what that type needs:

  • sentence-transformer — the model name, matching the model repository on Huggingface, so for example Alibaba-NLP/gte-base-en-v1.5.
  • openai and openrouter — the model id as the provider spells it, so text-embedding-3-large or qwen/qwen3-embedding-8b.
  • client-api — a client rather than a model. The selector lists the ready clients that expose an embeddings endpoint, and the model is whichever one that client has loaded.

Add new embedding

New sentence-transformer embeddings require a download

When you add a new sentence-transformer embedding and use it for the first time in the Memory agent, Talemate will download the model from Huggingface. This can take a while, depending on the size of the model and your internet connection. The hosted types — openai and openrouter — download nothing, and client-api uses the model your client already has loaded.

You can track the download in the talemate process window. A better UX based download progress bar is planned for a future release.

Editing an Embedding

Edit embedding

Select the existing embedding from the left side bar and you may change the following properties:

Trust Remote Code

For custom sentence-transformer models, you may need to toggle this on. This can be a security risk, so only do this if you trust the model's creator. It basically allows remote code execution.

Warning

Only trust models from reputable sources.

Device

The device to use for the embeddings. This can be either cpu or cuda. Note that this can also be overridden in the Memory agent settings.

Switching device without a restart (0.37.0)

Changing the device no longer requires restarting Talemate. The old model is released from ChromaDB's cache and any GPU memory it held is freed before the new device is applied, and the active scene's memory database is re-imported automatically.

Distance

The maximum distance for results to be considered a match. Different embeddings may require different distances, so if you find low accuracy, try changing this value.

Distance Mod

A multiplier applied to Distance when deciding whether a result is a match. The effective cutoff used during a search is Distance × Distance Mod, so this slider lets you fine-tune search sensitivity without changing the base distance.

  • Range: 0.1 to 2.0, in steps of 0.1.
  • Default: 1.0.
  • Lower values tighten the match (fewer, more relevant results).
  • Higher values loosen the match (more results, lower relevance).

Also tunable from the Context Database

The Context Database page exposes this same value as the Search Strictness slider, making it easy to adjust on the fly while testing searches. Changes made in either place are saved to the same preset.

Distance Function

The function to use for calculating the distance. The default is Cosine Similarity, but you can also use Inner Product or Squared L2. The selected embedding may require a specific distance function, so if you find low accuracy, try changing this value.

Fast

This is just a tag to mark the embedding as fast. It doesn't actually do anything, but can be useful for sorting later on.

GPU Recommendation

This is a tag to mark the embedding as needing a GPU. It doesn't actually do anything, but can be useful for sorting later on.

Local

Marks the embedding as running on your own machine. Turning it off hides the Device and GPU Recommendation controls, which mean nothing for a model served over an API. It is set for you when you add an embedding: only sentence-transformer is local.

System Prompts

App settings - Presets - System Prompts

This panel lets you override the global system prompts for the entire application for each prompt kind (Conversation, Narration, Creation, and so on). Per-client overrides live in the System Prompts section of each client's editor.

See System Prompt Overrides for the full reference, including:

  • Which prompt kinds exist and what they are used for.
  • How Normal and Uncensored variants are selected.
  • The pencil icon that marks entries with an active override (added in 0.37.0).
  • How to include the default prompt inside your override with {{ system_prompt }} (added in 0.37.0).