KoboldAI Lite

Welcome To KoboldAI Lite

Welcome to KoboldAI Lite!
Pick a UI Style to get started. You can always change it later in the Settings menu.

Classic UI

Aesthetic UI

Corpo UI

Alternatively, you can Load A Savefile or Quick Start Scenario

Context Data

Memory?Put the information you want the AI to always remember. It will be inserted into the top of every request sent to the AI.

Newline After Memory

Author's Note?Similar to Memory, but inserted near the end of the text instead of the start. A good way to control the mood/behavior of the AI.

Author's Note Template?A placeholder, will be inserted with the author's note replacing the <|>. You generally don't need to change this. A/N Strength?Controls how far back to insert the Author's Note. Notes injected closer to the end have a stronger effect.

?WorldInfo Groups can be used to segment data, e.g. entries for a specific place, person or event. Each entire group can be toggled on/off on demand.

Toggle Group

WI Insert Location ?Controls where the world info should be inserted

WI Search Depth ?Controls how far back in the text to search for World Info Keys

Case Sensitive Keys

Automatically search and include relevant snippets from a text document or history.

Enable TextDB

Search Includes Context History ? If enabled, the entire story/chat history is used as a searchable document

Num. Search Results ? Controls how many snippets to retrieve from DB

Nearby Text Amount ? Controls amount of nearby text included in the search query

Search Chunk Size ? Controls the size of each snippet being searched for

TextDB Storage?Paste as much raw text data here as you like. E.g. background information, reference documents, etc. This text will populate the database that will be chunked and searched by TextDB.

Usage Mode ?Story Mode is best for novel style writing. Adventure Mode is best for Interactive Fiction RPGs. Chat Mode is best for chat conversations with the AI. Instruct mode is for giving the AI ChatGPT styled tasks.

UI Style Select ?Select your preferred UI style, which affects text formatting and display. Some UIs are only available for specific modes.

Your Name ?The name the AI will see you as

AI Name ?Name of the person(s) you want to chat with. Multiple opponents can be specified, creating a group chat, separate their names using multiple lines.

Instruct Tag Preset ?Quickly select between common instruct tag formats. Different models are trained with different tags.

Instruct Tag Format ?Specify the exact text formatting used for instruct mode.

System Tag ?The sequence to start the system prompt

Sys. Prompt ?A fixed system message sent at the very start to guide the AI behavior. Usually NOT needed

User Tag ?The sequence to start an instruction prompt

Assistant Tag ?The sequence to start an assistant response

Multiline Replies ?Whether to allow multiple lines in AI responses (Chat/Adventure). Disable this if the AI starts generating rubbish.

Continue Bot Replies ?Allow AI to continue incomplete past responses, which can be extended by pressing submit again. Not recommended for newbies.

Chat Match Any Name ?In Chat Mode, match formatting against any name, instead of only specified chat names.

Chat PrePrompt ?Modifies the context, injecting tokens to improve chat quality for new chats.

Adventure PrePrompt ?Modifies the context, injecting tokens to improve adventure quality for new adventures.

Fix Alpaca Leakage ?Prevents leaking when Alpaca instruct format is used on models trained with bad/different formats.

Enable Markdown ?Allows the UI to use markdown formatting such as quotes and code blocks.

Enable LaTeX ?Allows the UI to render LaTeX within markdown formatting (Needs Markdown).

Trim Sentences ?Trims incomplete sentences in AI output.

Trim Whitespace ?Removes trailing whitespace in AI output.

Compress Newlines ?Compresses multiple newlines into one newline in AI output.

Inject Timestamps ?Injects timestamps into context (Chat/Instruct), allowing the AI to have a sense of time.

Inject ChatNames ?Appends chat names after every instruct tag, a hybrid chat mode.

Assistant Jailbreak ?Automatically injects a jailbreak message after every instruct query, to make the AI more likely to obey you.

Separate End Tags ?Allows using separate Instruction and Response End Tags, instead of combing them with the start tag. Not recommended for beginners.

Sampler Preset ?Pick from an easy selection of curated generation presets, or configure your own.

Context Size ?Maximum number of context tokens submitted to the AI. Must exceed max output tokens. Can be further increased by editing the textbox. Older models stop at 2048, newer ones can do 4096 or greater.

512

4096

Auto-Adjust Limits

Max Output ?Number of tokens the AI should generate. Higher numbers will take longer to generate.

512

Auto-Adjust Limits

Temperature ?Randomness of sampling. High values can increase creativity but may make text less sensible. Lower values will make text more predictable but can become repetitious.

0.1

Repetition Penalty ?Used to penalize words that were already generated or belong to the context (too high causes incoherence!).

Top-P Sampling ?Used to discard unlikely text in a nucleus sampling process. Lower values will make text more predictable but can become repetitious. Set to 1 to deactivate it.

Top-K Sampling ?Top-K Sampling. Discards all but the K most likely tokens. Set 0 to Deactivate.

Advanced Sampler Config ?These settings control alternative samplers configurations. They are inactive by default, you usually do not need to change them.

Top-A

Typical

TFS

Min-P

Pr. Pen.

Seed

Rp.Range

Rp.Slope

Smooth.F

DyTemp

EOS Token Ban ?Allow the End-Of-Stream (EOS) token and potentially other restricted special tokens to be generated.

Token Limit Multiplier ?Adds a multiplier to token estimations, useful if token counts are inaccurate. Influences text truncation. Higher values increases token allowance.

Sampler Order ? The order by which all 7 samplers are applied, separated by commas. 0=top_k, 1=top_a, 2=top_p, 3=tfs, 4=typ, 5=temp, 6=rep_pen

Mirostat ? Replaces your samplers with mirostat, an alternative sampling method. May not be available depending on backend, not supported on Horde.

Mode

Tau

Eta

Mirostat Not Supported

DRY (If supported) ?An advanced multi-token repetition penalty. Range controlled by Rep-Pen Range. May not be available depending on backend, not supported on Horde.

Mult.

Base

A.Len

DRY Not Supported

XTC ? Enables Exclude Top Choices (XTC) Sampling. May not be available depending on backend, not supported on Horde.

Threshold

Probability

XTC Not Supported

Grammar ? Grammar Sampling (KCPP) - Allows you to constrain output to fit specific structures. Resets grammar state every generation unless Retain is checked.

Retain

T.NSigma

Generate Images ?Use the AI Horde or a local KoboldCpp / Forge / A1111 instance to insert AI generated images into your story.

Detect ImgGen Instructions

Autogenerate

Save Images

Alerts ?Triggers alerts when a generation is completed.

Beep on Done

Notify on Done

Text To Speech ?Enable Text-To-Speech to have your story automatically read to you.

Narrate Both Sides

Narrate Only Dialog

Browser TTS Speed:

Voice Input ?Requires KoboldCpp with Whisper model loaded. Enables Speech-To-Text voice input. Automatically listens for speech in 'On' mode (Voice Detection), or use Push-To-Talk (PTT).

Suppress Non-Speech

Language

Voice Delay:

Style tags to use for generating images:
(E.g. Sketch, Realistic, Anime, 3D Render, Drawing)

Negative Prompt:

Number of Steps:

Cfg. Scale:

Sampler:

Resolution ? Changing resolution will affect the aspect ratio used to generate. This may impact quality or memory usage. :

Img2Img Strength ? Higher values lead to a more different image. :

Clip Skip:

Save Higher-Res ? This option will result in larger save files which may be slower. Changing this setting only applies to NEW images. :

Crop Images ? If enabled, oversized imported images will be cropped to fit. If disabled, images will be letterboxed instead. :

Insert Images as New Message ? If enabled, images will be added in a new turn message instead of the existing one. :

Search the Web for relevant information when using instruct mode
(Requires WebSearch enabled KoboldCpp)

Enable WebSearch

Use Multiple Passes ?Using this option will run a second LLM tool call to summarize context and create a more accurate search query. Slower but may be more accurate.

Multipass WebSearch Template ?The template used to generate the search query when multipass search is used

Retain Recent Searches ?Normally search results are only used once and then discarded. Enable this to retain search results for the previous 3 turns.

Extra Stopping Sequences ?Triggers the text generator to stop generating early if this sequence appears, in addition to default stop sequences. If you want multiple sequences, separate them with the following delimiter: ||$||

Logit Biases ?Specify a dictionary of token IDs to modify the probability of occuring.

Phrase / Word Ban (Anti-Slop) ?Prevents specific words or phrases from being generated, either modifying model vocab or by backtracking and regenerating when they appear. If you want multiple sequences, separate them with the following delimiter: ||$||

Regex Replace ?Allows transforming incoming text with regex patterns, modifying all matches. Replacements will be applied in sequence.

Thinking / Reasoning Tags ?Allows hiding or removing thinking tags.

Placeholder Tags ?Configure automatic substitutions for placeholders in text.

Token Streaming ?Use token streaming for partial responses. SSE is smoother but less well-supported. Poll is chunkier but more reliable. Not available on Horde.

Idle Responses

Never Escape HTML ?Avoids escaping any HTML tags, allowing HTML injections. Not recommended!

Render Sp.Tags ?If enabled, renders special tags like EOS and padding tokens. Not recommended.

Request Logprobs ?If enabled, request top 5 alternative token log-probabilities for each generated token. Incurs an overhead.

Autosave Session ?Autosaves your current story and settings on exit, reloads when you return

Embed Settings File ?Includes your current settings when saving or sharing your story

Rename Save File ?Prompts to input a different filename when saving file.

Show Advanced Load ?If enabled, allows you to select additional configurations during file load

Card Import Prompt ?If enabled, prompts the user to choose what mode to import a character card in. If disabled, automatically picks chat mode.

Run In Background ?Prevents the browser from suspending KoboldAI Lite by playing a silent audio track. This setting cannot be saved.

User Mods ?Allows you to load third-party user created mods (caution).

Custom CSS ?Allows you to load third-party CSS styles (caution).

Autoscroll Text ?Automatically scrolls the text window down when new text is generated

Unlock Scroll Height ?Unlocks the text viewport, allowing for infinite height without scrolling

Viewport Width ?Controls horizontal scaling of the viewport window

SidePanel Mode ?Displays the settings and context panels permanently on the sides, instead of as popups. Not available on Mobile displays.

Inverted Colors ?Inverts all colors, simple light mode

Background Img

Aesthetic UI customization panel

Background Style

Bubble Color:

System 🖌️

You 🖌️

AI 🖌️

Rounded Bubbles:

Color Background:

Min Height:

Horizontally-centered text:

Margin (px):

Padding (px):

Portrait Style

Portraits:

🖼️ Your Portrait

🖼️ AI's Portrait

Portrait Style:

(Reset Image)

User Portrait:

Size:

A/R:

AI Portrait:

Size:

A/R:

Show Names (Chat Mode):

Font Style

Font Size:

Customize:

Per-entity:

Style Text:

Colors:

text🖌️

"speech"🖌️

*action*🖌️

You:

text🖌️

"speech"🖌️

*action*🖌️

AI:

text🖌️

"speech"🖌️

*action*🖌️

System:

text🖌️

"speech"🖌️

*action*🖌️

Code blocks:

background🖌️

foreground🖌️

(Reset All Styles)

Style Preview

Quick Start - Select A Scenario

Automatically select AI model ? This option picks a suitable AI model based on the selected scenario. If no text model is currently selected, an appropriate one will be automatically picked for you.

Roleplay Character Creator

This tool is an easy way to create your own roleplay characters.
Alternatively, you can click here to import an existing Tavern Character Card.

Character Name:

Character Avatar:

Use Character Template ?Adds a character template to memory while creating the character instead.

Character Persona:

Describe Scenario:

Character Greeting:

After creating your character, you can edit it further at any time in the 'Context' memory window.

Save File / Load File / Export File

Slot Storage Option

Data Location

Selected Slot

Caution: Local Storage Slots are saved to a temporary cache and can be deleted by your browser. KoboldCpp remote storage will save data to a file in your KoboldCpp server. To avoid losing data, use the download file button.

Change Loaded KoboldCpp Config

Warning: This will terminate the current KoboldCpp instance and relaunch it with a new config.

If an invalid configuration is selected, the new server may fail to relaunch!

Really Start A New Session?

Unsaved data will be lost.

Keep AI Selected?

Keep Memory and World Info?

Advanced Load File

Select categories to import from saved file. Selected categories will be overwritten. Unselected categories will retain original values.

Main Story
Memory and Author's Note
World Info and TextDB
Tokens and Regex Settings
General Settings
Aesthetic Settings

Image Information

My Worker List

Name	Description	Uptime	Kudos	Maint.	Del.

Dynamic Temperature Wizard

Dynamic temperature is specified by a Temperature Value and a Temperature Range. Actual temperature is allowed to be automatically adjusted dynamically between (DynaTemp ± DynaRange).

For ease of use, a simple converter is provided here. Setting both values to the same temperature disables DynaTemp.

Minimum Temperature:

Maximum Temperature:

Temperature:

DynaTemp-Range:

DynaTemp-Exponent:

Paste Image From Clipboard

Share Story Import / Export