github SillyTavern/SillyTavern 1.13.0
on GitHub

latest releases: 1.13.3, 1.13.2, 1.13.1...
3 months ago

SillyTavern 1.13.0

Breaking changes

  • Chat Completion: The "Request model reasoning" toggle now controls just the visibility of the reasoning tokens returned by the model. To control the model reasoning request, use the "Reasoning Effort" setting. If unsure, "Auto" is the recommended option for most users. Please check the documentation for more details: https://docs.sillytavern.app/usage/prompts/reasoning/#reasoning-effort
  • CSS styles added to the "Creator's Notes" character card field are now processed the same way as styles in chat messages, i.e. classes are automatically prefixed, the external media preference is respected, and styles are constrained to the Creator's Note block.

Backends

  • Claude: Added Claude 4 models to the list. Added the extendedTTL parameter to extend the cache lifetime if using prompt caching. Added backend-provided web search tool support.
  • Google AI Studio: Reorganized and cleaned up the models list. Models which are redirected to other models are marked as such. Reintroduced the reasoning tokens visibility toggle.
  • Google Vertex AI (Express mode): Added as a Chat Completion source. Only Express mode keys are supported: https://cloud.google.com/vertex-ai/generative-ai/docs/start/express-mode/overview
  • Pollinations: Added as a Chat Completion source: https://pollinations.ai/
  • MistralAI: Added devstral and new mistral-medium models to the list.
  • OpenRouter: Synchronized the providers list.
  • llama.cpp: Enabled nsigma sampler controls. Added a min_keep setting. Disabled the tfs sampler as it is not supported by the backend.
  • Mancer: Enabled DRY and XTC sampler controls. Disabled the Mirostat sampler as it is not supported by the backend.

Improvements

  • Welcome Screen: Completely redesigned the welcome screen, added a recent chats display, automatic creation of a permanent Assistant, and the ability to set any character as a default Assistant. See the documentation for guidance: https://docs.sillytavern.app/usage/welcome-assistants/
  • Temporary Chats: Temporary chats can now be restored by importing a previously saved chat file.
  • Character Cards: Styles defined in the "Creator's Notes" field are now processed the same way as styles in chat messages and constrained to the Creator's Note block. Added a per-character setting to allow applying styles outside of the Creator's Note block.
  • Extensions: Added branch selection to the extension installation dialog. The branch can also be switched in the "Manage extensions" menu.
  • UI Themes: "Click-to-Edit" theme toggle is decoupled from the "document mode" style. Added an ability to set toast notifications position in the theme settings. Added a Rounded Square avatar style.
  • Style tags defined in greeting messages will now always be applied, even if the message is not rendered. Use the "Pin greeting message styles" user setting to control this behavior.
  • World Info: Added per-entry toggles to match entry keys with the character card fields.
  • Chat Completion: Added source-specific Reasoning Effort options: Auto, Minimum, Maximum. The "Request model reasoning" toggle now only controls the visibility of the reasoning tokens returned by the model.
  • Chat Completion: "Prompt Post-Processing" can be used with any Chat Completion source. Added "Merge into a single user message" option to the post-processing settings. Tool calling is not supported when using Prompt Post-Processing.
  • Chat Completion: Added a toggle to control the link between Chat Completion presets and API connections. When enabled (default), API connection settings will be bound to the selected preset.
  • Prompt Manager: Added an indication of where the prompts are pulled from. Added an ability to set priorities of prompts on the same injection depth (similar to World Info ordering behavior).
  • Text Completion: Added a Post-History Instructions field to the System Prompt settings.
  • Text Completion: Added GLM-4 templates. Fixed Lightning 1.1 templates. Pygmalion template merged with Metharme template.
  • Advanced Formatting: Non-Markdown Strings do not automatically include chat and examples separators anymore. Use {{chatStart}},{{chatSeparator}} value to restore the classic behavior.
  • Backgrounds: Video backgrounds can now be uploaded with automatic conversion to animated WebP format. Requires a converter extension to be installed: https://github.com/SillyTavern/Extension-VideoBackgroundLoader
  • Server: Added a --configPath command line argument to override the path to the config.yaml file. Missing default config entries will be added even if the post-install script is not run.
  • Tags: Added an ability to hide tags on characters in the character lists.
  • Various localization updates and fixes.

Extensions

  • Image Generation: Added gpt-image-1 model for OpenAI. Added {{charPrefix}} and {{charNegativePrefix}} global macros.
  • Image Captioning: Added Pollinations as a source. Added secondary endpoint URL control for Text Completion sources. Fixed llama.cpp captioning support.
  • Vector Storage: Added embed-v4.0 model by Cohere.

STscript

  • Added /test and /match commands to perform RegEx operations on strings.
  • Added raw=false argument to control the quotes preservation of the message-sending commands (e.g. /send, /sendas).
  • Added /chat-jump command to quickly scroll to a message by its ID.
  • Added a name argument to the /sys command to set a name displayed on the message.
  • Added /clipboard-get and /clipboard-set commands to read and write to the system clipboard.

Bug fixes

  • Fixed vectors generated by KoboldCpp not being saved correctly.
  • Fixed group chat metadata being lost when renaming a group member.
  • Fixed visual duplication of Chat Completion presets on renaming.
  • Fixed sending a message on Enter press while IME composition is active.
  • Fixed an edge case where the Continue suffix was not correctly parsed in instruct mode.
  • Fixed compatibility of tool definitions with the DeepSeek backend.
  • Fixed xAI selected model not being saved to presets.
  • Fixed a server crash on extracting corrupted ZIP archives.
  • Fixed "hide muted sprites" toggle not being preserved per group.
  • Fixed logprobs token reroll when using auto-parsed reasoning.

Community updates

New Contributors

Full Changelog: 1.12.14...1.13.0

Check out latest releases or
releases around SillyTavern/SillyTavern 1.13.0

Don't miss a new SillyTavern release

NewReleases is sending notifications on new releases.

Get notifications