- added support for OpenAI's Vision capabilities (thanks to @Anajrim01)
- migrated from localStorage to indexedDB because 5MB won't be enough to save images
- fixed token calculation (thanks to @Anajrim01)
- improved Markdown rendering for tables and paragraphs (thanks to @Anajrim01)
- replaced
snarkdownwithmarkdown-it. This should make code highlighting in completions more reliable, supports more syntaxes like LaTeX and fixes list numbering in completions. (thanks to @Anajrim01) - auto-generate titles after you have typed the first message (thanks to @Anajrim01)
- added Mistral (Cloud) and Meta/Llama3 (via Groq) models to "bring your own API key" mode. We are waiting for an Azure fix to serve them using Wingman Pro. Stay tuned!
- deprecated
gpt-4-turbo-previewin favor of the currentgpt-4-turbomodel - removed larger gpt-4 models because it doesn't make sense to use them without file upload or image capabilities.
- Redesigned SlickGPT to make more room for the actual chat and use the available screen-space better:
- removed footer
- lowered fontsize
- title and action bar now scroll and aren't pinned anymore
- removed language hint
- shrank chat input and context input
- shrank dashboard tiles
- added model to dashboard tiles
- reworked mobile viewport as well
- added new gpt-4o model
- deprecated gpt-4-turbo-preview model in favor of gpt-4-turbo
- SlickGPT now offers an optional "Pro" plan. If you choose this, you can use our Azure cloud infrastructure to access all the OpenAI models and settings without worrying about your own API key or usage limits. Yes, it's unlimited. We added a bunch of environment variables that you have to specify in your
.envfile to make this work. Check the.env.examplefile for details. You can leave all the values empty but you have to declare the variables if you want to self-host. - The Svelte endpoint layer for
/askand/suggest-titlewere removed. The client now communicates with Open AI (using your own API key) or our Azure backend (if you are on the Pro plan) directly. This should simplify self-hosting as you no longer need serverless/Edge functions to execute prompts.
- incomplete code blocks in completions should no longer overflow their containers
- fixed dashboard layout on smaller screens/resolutions
- add CommandBar to search and jump to chats (thanks @th8m0z)
- integrate Moderation API, can be disabled in .env file (thanks @th8m0z)
- upgraded to new OpenAI 0125 preview models and model aliases
- Pressing ENTER in the "Edit Chat Title" modal now saves the title instead of suggesting one via GPT
- Fixed gpt-4-turbo maxToken length
- Links to our new Discord and Patreon were added. Feel free to remove them in your custom instance!
- Messages with code now display a "Copy Code" button onHover
- upgraded all client libs, including Svelte4, Skeleton 2.0 and OpenAI 4
- Fixed an issue in Firefox that made the ChatInput appear too small in height
- Maintenance: Update client libs, use Svelte 4 etc.
- add new enironment variable PUBLIC_DISABLE_TRACKING. Set this to
trueto prevent tracking actions.
- improved performance in large chats by initializing the tokenizer only once (thanks @Schroedi)
- minor bugfix: resize ChatInput on every input event (thanks @Schroedi)
- Opened but unclosed code tags in completions are now auto-closed automatically. This should make streaming (and partial) results more readable. Thanks to @Arro for the suggestion!
- cancelled completions are now added to the chat history as well. They might be malformatted or incomplete, but they are there.
- only autoscroll if user is at the bottom of the chat. Thanks to @thenbe for the suggestion and PR!
- Messages are now editable. This creates a branch in the chat history and preserves your old prompts and completions. You can continue writing in any branch, just like you are used to from the OpenAI client. This change requires a (in-place, client-side) migration of the chat data structure, so you might have to wait a few seconds for each chat to load for the first time after the update.
- Suggest Title button now shows a loading spinner and is disabled while loading
- Chat titles are now also shown on mobile, along with better styling
- Toast messages (especially primary) look a bit better now
- use svelte-legos to replace some (worse) custom logic
- textareaAutosizeAction instead of
svelte-autosize - clickToCopyAction instead of custom
navigator.clipboardlogic
- textareaAutosizeAction instead of
- long Chat titles no longer break the layout
- use @inqling/svelte-icons instead of inline SVGs
- added a little surprise. Can you spot it? 👀
- Auto-Suggest chat titles: When you leave an untitled chat, a new Modal will appear and allow you to set the title or let
gpt-3.5-turbosuggest one for you (based on the context and first prompt and completion). You can opt-in to always do that by default. - Chat title moved from Chat Settings Modal to a new Edit button next to the title (that will also open the Modal mentioned above)
- Chat messages can now be deleted (both user and assistant) by clicking the x button next to them
- You can now cancel completions while ChatGPT is generating
- OpenAI API key is now masked in the Chat Settings modal once you entered it
- Added a new Insert code button next to ChatInput that appends Markdown code tags to the input. Use this for code snippets, SVGs etc. that you want to feed to ChatGPT.
- Empty chats (just created, no context or messages) will be deleted automatically when you close them.
- fixed code highlighting in shared chats
- If you change the model in the Chat Settings, you can set your new model as default for all subsequent chats
- added New Chat button next to context which will create a new chat with the same context and settings
- debounce token calculation on input change to avoid laggy UI (thanks @Ratcha9)
- fixed token count being unreadable in light theme
- remove text blink while receiving a completion from ChatGPT