Release Notes

Desktop:

• Add Ollama v0.9.5.

• Add 10 new LLMs which you can one-click download.

Web:

• Fix a bug with Proxy LLM URLs not working.

Web & Desktop:

• Add lorebooks to user personas.
• Support exporting characters as Tavern Card V2 PNGs.
• Allow uploading multiple gallery images at once.
• Allow uploading multiple character card PNGs or Backyard AI characters at once.

Desktop:

• Add larger context sizes.
• Fix bug when clicking authors on the leaderboard page.
• Fix bug where the loaded Ollama models were not fetched.
• Add button to the help menu to show the update logs.

Web & Desktop:

• Add a way to import Backyard AI characters.
• Set AUTHOR_NOTE_MAX_CHARS to 2000.
• Prevent publishing stories which are empty.

Desktop:

• Update so that if you have a Ollama version downloaded but don't have it selected, we will auto-select it on app start.

Web:

• Remove WebLLM support, i.e. running WebGPU LLMs in the browser. It just wasn't good enough, but maybe some day we can add it back.

Web & Desktop:

• Fix a bug where the image model and voice select modals were not visible.

Desktop:

• Fix a bug in the updater - I think it should be better than it used to be, but would love to hear if not!

Web & Desktop:

• Fix some UI bugs when uploaded gifs / webp images were not squares.
• Fetch 30 characters at a time when browsing the characters page instead of 20.

Desktop:

• Improve how we kill the Ollama process on app close.

Web & Desktop:

• Add text-to-speech! This is available to Ultimate users only, and I'd love any feedback on how you like it.
• Update the proxy LLM code to strip `/chat/completions` or `/completions` from the end of the URL. Before only a URL like "http://localhost:1234/v1" would work, not "http://localhost:1234/v1/chat/completions".

Desktop:

• Fix a crash where when using default background images in offline mode, the app would crash. Now we cache the default background images locally.

Web & Desktop:

• Fix bugs with some of the image generation models, and improve errors when they fail.
• Add Google Imagen 4 Fast and Minimax Image 01 to the image generation models. These are quite high quality SFW models.
• Fix a bug when you tried to create a group chat with only one character.

Desktop:

• Fix a bug with the new Ollama update.
• Add Ollama v0.9.1, 0.6.8, 0.5.13, and 0.4.7. Hopefully this will be useful so you can test older versions if you're running into issues with the GPU not being used.

Web & Desktop:

• Add image generation to stories.
• Add lorebooks to stories.
• Allow GIF and WEBP images.

Desktop:

• Updated Ollama and ability to choose your Ollama version.
• Fix bug where the Ollama process was not getting killed on app close.

Web & Desktop:

• Add a Release Notes page.
• Improve the UI of the author's page. This is in preparation for a big overhaul coming: optional user-login.

Web:

• Generate opengraph images for character and story URLs.

Web & Desktop:

• Add a button to the character editor to generate images, and then set them as the icon image, background image, or gallery images.

• Fix a bug where changing the image model could break the selected LLM model.

• Improve error messages when the cloud LLM fails.

Web & Desktop:

• Add a new UI toggle to the `My Characters` page - when on, we use cards like on the `Characters` page.

• Remember the sort setting you had selected last on the `My Characters`, `My Stories`, and `Chats` pages.

• Fix bug where alternate first messages were not being imported with character card PNG imports.

• Fix missing alternate first messages during character upload and duplication.

• Set the max length of the lorebook to 10,000 chars.

• Fix bug where the `Edit` panel wasn't showing up when chatting with characters which you own.

• Add way to sort characters A->Z or Z->A in the `My Characters` page.

• Add more random image prompts for you to try out.

Web & Desktop:

• Scroll to the bottom of the chat on conversation start.

• Make Mistral-Nemo Instruct 12B the default free cloud chat model.

• Fix a bug with the chat auto-saving when you edit a message.

• Fix a bug where newlines did not show up in the chat, all the text would be wrapped.

Desktop:

• Update so that your zoom level is remembered between app restarts.

• Add new app icons which have a white background instead of a transparent background.

Web & Desktop:

• Allow customizing chat message UI - you can now customize with chat bubbles, background colors, font sizes, and text colors.

• Fix bug with the chat input text box having flickering when you type a lot.

• Allowing changing the image model on the chat page.

• Show what image model and LLM you are using in the top right corner of every page (it was missing from the Create Image page).

• Fix bug where if you imported a new Lorebook JSON, it would erase old Lorebook entries.

• Fix bug where when you manually edited a message, it would automatically auto-save your chat.

Web & Desktop:

• Fix some bugs with image generation.

Web & Desktop:

• Adds a new option to have chats auto-save. Sorry this took so long!
• Adds the ability for non-paid users to still see the Chats page - this way you can stop your subscription and still see your chats.

Web & Desktop:

• When browsing characters, adds the ability to completely hide characters with certain tags - you can configure this on the Settings page.
• When browsing characters, adds the ability to filter against certain tags. Right click on a tag to exclude it from searches.
• Improve the story writing experience - in addition to Regenerate, Expand, Shorten, and Randomize, now we have a Quick Edit button which you can prompt. Also adds keyboard shortcuts for all of these.
• Adds a Gallery to characters - now you can upload up to 10 images alongside your character Avatar and Background Image.

Web & Desktop:

• Fix a bug where cloud LLMs would repeat themselves - this was a major bug that should improve the quality of longer cloud LLM chats significantly.
• Adds many more image generation models! These are faster, more stable, and better quality. You can see the full list here: https://www.hammerai.com/plans
• Adds "Mistral-Nemo Instruct 12B" as a free cloud chat model, moves "Mythomax 13B" to Starter, adds "Mistral Small 3 24B" to Advanced, and adds "WizardLM 2 8x22B" to Ultimate. Hope you like the change!

Web:

• Adds Stories to the leaderboard page! Now you can compete to be the best story author: https://www.hammerai.com/leaderboard
• Add image generation models to status page - now you can see how long they take, and their success rates: https://www.hammerai.com/status

Web & Desktop:

• Add alternate first messages to characters! Now you can create a few options, and then select which one to start with.
• Add more theme colors.

Web & Desktop:

• Add group chats! Would love to hear how you like it. I know there are still issues with characters responding with "<their name>: ..."
• Add Author's Note! This lets you write a note which is the last thing the LLM reads. You can use this to direct the chat in a certain direction, or improve memory.
• Add Character Visibility! This lets you control whether the character info of characters you upload to the cloud is visible to other users. If you set to Private, other people can chat with your character, but not know about the character info / Lorebooks / etc.
• Add more generation settings to Story Mode - you can now control how much of the story is fed to the LLM. The prompt is also improved.
• Improve image generation - now we use a polling mechanism to check if the image is ready, which is much more reliable.
• Add new tags!
• Fix a bug where if your subscription was refunded, you would not be able to chat for the rest of the month.

Web:

• Add a new Settings page like we have on desktop. You can use this to manage Background Images, User Personas, Author Mode, and more.

Web & Desktop:

• Add 10 new cloud image generation models. I'd love to hear if there are any in particular you like!
• Add a Leaderboard page! This shows the top character creators. Soon it will also show the top story authors.
• Improve the My Characters page UI - now there are three columns: Favorite Characters, Local Characters, and Cloud Characters. You can also filter by tag.
• Improve the image generated within chats.
• Fix a bug where when writing stories, if the story got too long, the LLM wouldn't be able to come up with new content. Now we only feed the last 500 words to the LLM so that we don't fill up the context.
• Improve the UI of reading stories - the text has less line spacing, you can configure the story text size, you can rate stories, and you can see how many views and the rating of every story.

Web:

• Add a new docs section! I hope you find this useful, much of the content was written by @May - a million thanks to him!! If you have any suggestions, please let us know. You can view it here: https://www.hammerai.com/docs
• Improve the character editor UI on mobile devices.

Web & Desktop:

• Add a new Proxy Model mode! This lets you use any OpenAI compatible LLM API instead of just HammerAI managed APIs. I've added some notes about how to use Featherless to the app, and would love to hear if you use anything else! Once we have documentation working on the site, I'll be adding guides for how to use different proxy providers.
• Improve the performance and stability of the cloud-hosted Jamet MK.VBlackroot 8B model.
• Add a confirmation window when deleting user personas and background images.
• Small UI fixes: remove duplicate "Fill in the rest for me" button, remove duplicate "Display Name" label, and improve the character description box used on the My Characters and Chats pages.

Desktop:

• Start building a new MSI Windows package.

Web & Desktop:

• Add button to import lorebooks from JSON. You can find some to use at: https://www.characterhub.org/lorebooks
• Add ability to sort characters and stories alphabetically.
• Add button that will use a LLM to improve your image generation prompt.
• Improve the "Impersonate" LLM prompting when chatting with characters.

Web:

• Fix bug where character's first message did not show up on mobile devices.
• Fix bug where the message box would look bad on mobile devices when the page first loaded.
• Improve accessibility by adding some missing aria labels.

Web & Desktop:

• Improve chat page UI on small screens.
• Improve performance of navigating around the app.
• Fix a bug when generating images.

Desktop:

• Add a RPM package for RedHat-based Linux distributions to the app release.

Desktop:

• Add new models: Rocinante 12B, Mistral Nemo Inferor 12B, Mistral Nemo Mag Mell R1 12B, Mistral Nemo Celeste V1.9 12B, Tiger Gemma 9B, Big Tiger Gemma 27B, Magnum V1 72B, Cydonia 22B, DeepSeek-R1 8B, DeepSeek-R1 14B, DeepSeek-R1 32B, and DeepSeek-R1 70B. I suggest you try out Mistral Nemo Mag Mell R1 12B, as I found it quite good. In the coming weeks we'll do a new poll to see which is your favorite, and which we should make the default.

Web & Desktop:

• Fix bugs with image generation: removes negative prompt and guidance from the cloud flux image generation model, as they are not actually used.
• Add image generation to within the chat window! I would love to hear how this works for you, and what settings you'd like added here to help you generate better images during chats.
• Significantly improve the speed and stability of the cloud LLM generation for Jamet MK.VBlackroot 8B, Psyonic Cetacean 20B, and Midnight Rose 70B. If you had a bad experience with these models, please try again.

Web & Desktop:

• Release image generation! To start it is just cloud image generation using Flux 1 Schnell, and is only available to Advanced and Ultimate users. But in a follow-up release I'll be adding local image generation, which will be free.

Desktop:

• Allow you to set larger context sizes on the Settings page.

Web & Desktop:

• Fix UI issues in the sidebar on smaller screens.,

Desktop:

• Fix a bug where when you first downloaded the app to a new computer, you may not be able to create local characters.

Web & Desktop:

• Show the number of cloud messages sent to each character - note this doesn't include local messages (i.e. using Ollama / WebGPU).
• Add a button to character cards (which have an author) to go to the author's profile.

Desktop:

• Fix a bug where uploaded characters would not have their Author Note.

Web & Desktop:

• Adds themes! We'll start with a light and a gold theme, but will add more soon. Also, please note there are still a few bugs here, but I'll fix in the next release.
• Adds an Author Note to characters - you can fill this in to write something to users who use your character.
• Adds a Display Name to characters - this will be shown instead of the character name in the character browser, but the AI will not know this.
• Fix a bug where lorebook entries did not get downloaded when you downloaded a character.

Web:

• Add a new status page to track how well cloud models are performing - right now they are horrible! I'm really sorry. I will be investigating and improving things, which should be much easier now that I actually track things. https://www.hammerai.com/status

Web & Desktop:

• Adds themes! We'll start with a light and a gold theme, but will add more soon. Also, please note there are still a few bugs here, but I'll fix in the next release.
• Adds an Author Note to characters - you can fill this in to write something to users who use your character.
• Adds a Display Name to characters - this will be shown instead of the character name in the character browser, but the AI will not know this.
• Fix a bug where lorebook entries did not get downloaded when you downloaded a character.

Web:

• Add a new status page to track how well cloud models are performing - right now they are horrible! I'm really sorry. I will be investigating and improving things, which should be much easier now that I actually track things. https://www.hammerai.com/status

Web & Desktop:

• Adds two new account types: Advanced and Ultimate. Advanced accounts get access to Psyonic Cetacean 20B with a max context window of 8,192 tokens, and Ultimate accounts get access to Midnight Rose 70B with a max context window of 30,720 tokens. If you are a Pro user, you can upgrade your account to Advanced or Ultimate on the Gumroad portal.
• Increase the length of messages you can send to characters.

Web & Desktop:

• Adds support for cloud LLMs! These will be run on a remote server which HammerAI manages, and will eventually let you use larger models that you may not be able to run locally. Free users get unlimited messages with Smart Lemon Cookie 7B, and Pro users get unlimited messages with Jamet MK.VBlackroot 8B.
• Add a link to the Author page to the chat page Character tab.

Web:

• Adds support for Regenerate, Continue, Write for me, and Undo in character chats.

Web & Desktop:

• Adds support for Lorebooks - this lets you add dynamic memory to your characters.

Desktop:

• Improves "Write for me" feature in Write Story mode.

Web & Desktop:

• Fixes a bug where Modals would not have the confirm/cancel buttons.

Web:

• Fixes a crash when WebGPU is not available.

Web & Desktop:

• Fixes a bug in the verification pop-up.
• Fixes a bug where you could not clear your author name.

Web & Desktop:

• Fix a bug where updating character settings with the input box would not work.
• Add a pop-up to explicitly verify your age before chatting or writing stories.
• Add moderation around self-harm - this check runs locally and does not send any data to any server, but will prevent you from sending messages that contain references to self-harm.

Web & Desktop:

• Fix a bug where the first message did not have {{user}} replaced with your name.
• Fix a bug where updating the character / LLM settings would not be persisted.
• Refactor the dropdown for model selection.

Web:

• Update the LLM to use a newer version of WebLLM and use a webworker when processing. The result is that the LLM runs much faster, is more coherent, and is more stable.

Desktop:

• Adds "Continue generating" and "Write for me" buttons to the chat page. "Continue generating" will generate more text, and "Write for me" will write the text for you.
• Adds "Make longer", "Make shorter", and "Randomize" buttons to the Write Story page when you highlight some text. "Make longer" will make the text more verbose, "Make shorter" will make the text more concise, and "Randomize" will rewrite it and introduce new ideas.

Desktop:

• Adds background images to characters. You can now set a background image and opacity on your characters, which will be shown in the chat. Also adds a toggle to the Chat page to override the character background image with your own custom local background image.
• Fix the tooltip jitter bug on the characters page on smaller screens.

Web & Desktop:

• Adds a button to delete the background image for a character.
• Fixes sliders being laggy.

Web:

• Adds support for background images and user personas.

Desktop:

• Adds forward and back buttons to go between pages.

• Makes several small UI fixes for smaller screens.

• Updates the UI: now each page has a top bar with a draggable region.
• Adds a new button to the create character page which will automatically generate the rest of the character info for you.
• Adds search and sort to the Models page.

Hi all, version 0.0.149 is live! This release includes:

• Adds a new Featured tag - if you want your character featured, post in ⁠🧚︱characters-sfw or ⁠🔞︱characters-nsfw and share the URL to it with me! I'll also be adding characters to this as I review them.
• Localizes content that was missing from the previous few updates.

• Adds Download and Copy Character URL buttons to the chat sidebar.

• Adds chat background images! You can now choose background images which you'll see during a chat. Eventually I'll also add this to individual characters, but for now they are just global, local settings.
• Fixes a bug where you could not read stories.
• Adds auto-scroll to stories.
• Split up the Story settings from the Publish settings on the Write Story page.
• Add an extra-large character image size - on the right sidebar just click on the image to make it bigger.

• Adds chat background images! You can now choose background images which you'll see during a chat. Eventually I'll also add this to individual characters, but for now they are just global, local settings.
• Fixes a bug where you could not read stories.
• Adds auto-scroll to stories.
• Split up the Story settings from the Publish settings on the Write Story page.
• Add an extra-large character image size - on the right sidebar just click on the image to make it bigger.

• You can now add images to stories.
• You can now add a story category, your own tags, and a rating.
• Stories are now saved locally automatically.
• You can publish stories to the cloud, which will make a copy on HammerAI servers. You can then unpublish them (which keeps your local copy) or delete them (which deletes your local copy).
• You can browse stories written by other people.

Desktop:

• Adds the Story Title, Description, Genre, Style, and Synopsis to the Story Mode. This will help the AI generate better stories for you. I also added an AI-generator where all of these will be generated by the AI - very meta!

Desktop:

• Adds a new mode: Story Mode! With Story Mode you are given a blank slate to write a story using AI. It's pretty early and there is still a lot I want to add, but I'd love to hear how it works for you!
• Allows you to turn off spellcheck in the app.
• Allows you to right-click on spellcheck errors to get suggestions. This happens locally and does not send any data to any server (https://www.electronjs.org/docs/latest/tutorial/spellchecker#does-the-spellchecker-use-any-google-services).

Web & Desktop:

• Expose the system prompt, temperature, repetition penalty, topK, and topP as settings on each character, and also let you override these with your local global settings.
• Add user personas - now you can have multiple user personas with a name, description, and profile picture. You can also easily switch between them. Note that switching will only affect future messages, not past messages.
• Add Author Mode, as a way to publicly share which cloud characters are yours. Note that this is opt-in - if you do not enable this for each character, it will stay 100% anonymous.
• Allow updating of character images.
• Adds a new "Sort by most recently upvoted" option to the characters page
• When browsing local characters hide the left and right page buttons.
• Fix bug with inserted example message not having hashtags.
• Improve the UI of the sorters.

Desktop:

• Expose the default LLM as a setting on each character, and also let you override this with a local global setting.
• Fix an issue where the server sometimes would fail to start-up. We now retry once in that case which usually fixes the issue.
• Add smart lemon cookie q3 and q5.
• Fix a bug where you couldn't stop message generation.
• When downloading a custom Ollama model you can paste in ollama run foo and we will download foo.
• Let's you see and unload models that have been loaded into RAM

Web:

• Adds a blog

Web & Desktop:

• Allow even longer character fields to be uploaded to cloud.

Desktop:

• Adds Ukrainian translations - if you have any other languages you would like, let me know!
• Fixes a bug where we were logging to the App Logs during model loading.

Web & Desktop:

• Improves the chat page UI and allows you to edit your characters as you chat.
• Fixes a bug where the messages would not scroll as you chat.

Desktop:

• Updates from using llama.cpp directly to using Ollama, which wraps llama.cpp. This will improve stability and GPU usage, leading to faster responses for you! You will need to re-download your models, but I think it will be worth it.
• Improves the Windows UI to look more native.

Web & Desktop:

• Updates the UI to use a sidebar instead of a header + a footer.
• Makes a few small UI improvements to the chat page.

Desktop:

• Fix a few small offline mode bugs.
• Adds internationalization support - I added several languages using AI translation. If you know any of them and see any issues, please let me know! And if you have any other language you'd like to see, please let me know and I'll add it.

Web & Desktop:

• Fix bug where you couldn't upload cloud characters.
• Fix character card website link (it used to be to booru, now it's to character hub).
• Add links to our TikTok (https://www.tiktok.com/@hammerai.com), Reddit (https://www.reddit.com/r/HammerAI), and Twitter (https://x.com/HammerAIDotCom) pages.

Web:

• Expose the My Characters and My Chats pages on web.
• Allow you to save chats on web.
• Change from using local storage to using IndexedDB to store your characters. This should improve performance as you save more characters and chats.
• Fix accessibility issues and load more images from a CDN instead of the server.

Web & Desktop:

• Allow you to select multiple tags at once. If you select A and B, we will only show characters that have both A and B tags.

Desktop:

• Splits up data we store between three JSON files: 1) config.json which stores your settings, 2) characters.json which stores your characters, and 3) conversations.json which stores your chats. This should improve performance if you have many characters or chats.

Desktop:

• Adds a new Experimental llama.cpp backend. This has several months of llama.cpp fixes in it which should increase the stability, quality, and speed of the LLM generation. I've tested in on Mac and Windows with a NVIDIA GPU, but not on Linux or Windows ARM. Would love to hear how it works for you and if you have any issues. You can enable it in the Settings page under Backend Version -> Experimental.
• Fixes the issue where long conversations would become incoherent - there are still issues with the model forgetting parts of the previous conversation, but now it won't get stuck like it used to.

Desktop:

• Add the ability to name conversations - hopefully this will help you keep track of them more easily!

Web & Desktop:

• Fix a bug where a local character creation timestamp would get overwritten when you edited the character

Desktop:

• Fix a bug where the model wouldn't be downloaded correctly
• Add a button which will let you download a model immediately upon downloading the app
• Update how we package the llama.cpp binaries - if you see something break because of this, please let me know!
• Fix a bug in our token counting logic - this may help avoid model incoherence during longer conversations
• Improve error handling during model loading - we now will display better error messages and avoid infinite loading when things go wrong (in some cases)

Desktop:

• Allow users to change where the LLMs are downloaded to
• Add a Discord link and TikTok link (https://www.tiktok.com/@hammer_ai) to the footer

Web & Desktop:

• Adds toasts to confirm when you've taken an action (e.g. updating or deleting a character).
• Adds confirmation modals which will ask you to confirm before you delete a character or chat.

Desktop:

• Fix a bug where in the chat UI {{user}} would not be replaced with your name.

Desktop:

• Only apply the new GPU Layers setting to Macs and Windows machines with CUDA, as it was breaking responses on other machines. In a follow-up we'll investigate why this is happening and use a better solution.

Desktop:

• Fix a bug where the chat would run very slowly after the 0.0.60 release - we were reading from the JSON file each time you or the LLM typed anything, which was very slow.
• Fix a bug where if you cancel your Pro subscription you would immediately lose access - now you have access until the end of your billing period.
• Expose two new settings:GPU Layers: This offloads more processing on your GPU. Before we had it set to 1 which means we didn't use the GPU to the full possible extent, but now we set it to 50. But you can try setting it even higher if you have a good GPU. In a follow-up we can automatically calculate this for you, but for now it's manual.,Max Response Tokens: We set this to 256 today - if you find that your chat is cut off, you can increase this number.

Web & Desktop:

• Fix sorting of local characters and favorited characters on the Characters page.

Desktop:

• Move local characters above cloud characters on the Characters page.
• Remove the local tag from characters before you upload them (this was an internal bug where we would accidentally mark the characters you upload to the cloud from your local characters as local characters, even when they become cloud characters)

Web & Desktop:

• Add a search bar to the Characters page.
• Move more delete buttons to the right side so you don't click on them by accident.

Desktop:

• On the My Characters and Chats pages you can sort by creation date, update date, and message count.
• On the Settings page you can add a profile picture which will be shown while you're chatting.
• You can now download characters from the Characters page to be local characters, which you can edit.
• You can now upload your local characters to be cloud characters. Doing this will make them public (after they have been reviewed by moderators) and allow others to chat with them.
• You can now duplicate your local characters if you want to try out different versions of them.

Web & Desktop:

• On the chat page moves the Delete button to the right side.
• On the Characters page you can now Sort by most recent, Sort by oldest, Sort by most liked, or Sort by least liked.

Desktop:

• Adds Smart Lemon Cookie 7B, Fimbulvetr v2 11B, and Chaifighter v2 20B models.

Desktop:

• Fixes a bug where the character image would be deleted when you update the character.

Desktop:

• Adds saved conversations. I hope you all don't hate me, but I added this as a Pro feature. The goal for me is not to make money, instead I want to fund paying other developers to improve the app. I have too much going on in my life to give this app the love is deserves, but the foundation is very solid, so I think that even being able to pay someone for a few hours a week will really lead to nice results for you all. I hope you understand If you want to try it out, check out the new ✨ Pro tab in the header. Payments are handled through Gumroad (https://hammerai.gumroad.com/l/store), and you'll get a license key to enter after paying.
• When creating local characters we used to store the character image as a base64 string in config.json. This led to long load times when you had many characters. Now we store the character images under /images and reference them in config.json. This should lead to much faster load times. Unfortunately, this means existing characters will remain slow. I'm sorry about that - my suggestion is to re-create them again.

Web:

• Removes models where the download link doesn't work,
• Update to a new version of Web LLM. It is faster than before which is great, but we lost the uncensored Mistral models which we had before. I will work on getting these back soon. Alternatively, if anyone wants to contribute to this project, you could build a q4f32_1 version of a 7B uncensored model and put it on HuggingFace! You can follow the instructions here to do so (and let me know if you're interested, I can help walk you through): https://llm.mlc.ai/docs/compilation/convert_weights.html,,
• Add a new status page where you can go to see if something is wrong with the web model downloads: https://www.hammerai.com/status,
• Fix a UI bug when reloading a model.

Desktop:

• Improves the logging when you chat with characters. If you have issues with an infinite chat loading experience, please let me know and include the logs from Help -> View App Logs. Hopefully with these new logs I can understand where the issue is.,
• Add the Intel build to the download page. This is a beta build, so please let me know if you have any issues with it.

Web & Desktop:

• Updates so that when creating characters locally there is no limit on the description, personality, etc.

Desktop:

• Enable infinite chat sessions! We do this by following the examples/chat-persistent.sh example in llama.cpp, which shows how to remove older messages if we predict that we'll exceed the context length. This is required because we set n_keep to -1 so we hit the context size we retain all tokens from the prompt (and thus, before this change, any new messages would then be ignored).
• Adds the ability to regenerate the last system message, to delete older messages, and to edit the text of any message. Please let me know if the behavior you see here is not what you expect.
• Fixes a bug where if you had a space in your path you would not be able to chat. There may still be other issues here, so please let me know if you have any problems.
• Changes the auto-update check to be every 20 minutes instead of every 10.
• Adds a setting in the Settings page to configure the context size.

Desktop:

• Adds local characters.

Desktop:

• Adds the ability to delete characters.
• Removes the need to download CUDA on NVIDIA machines, instead we ship the app with the CUDA libraries and use them if Enable GPU is true.
• Use CLBlast on AMD machines if Enable GPU is true. Like on NVIDIA machines, you should not need to download anything as we ship with the CLBlast library.
• Decrease the size of the Windows package from 806 MB to 393 MB.
• Build the Ubuntu package on a Ubuntu 20.04 machine and modify the permissions of the llama.cpp binary (add execute permissions).
• Build a flatpak version for Linux users.

Desktop:

• Adds information about GPU and AVX acceleration to the Chat page.
• Adds more debug logging about GPU support (you can find this under Help -> View App Logs).

Web & Desktop:

• By default we show only SFW characters page on the home page, but we will use local storage to remember if you want to show all characters instead.
• Increases the maximum number of characters in the character personality from 2000 to 3000, and the maximum number of characters in the first message from 400 to 500.

Desktop:

• Adds beta support for Windows! There may be bugs here, so please let me know any issues (and attach the logs you find in Help -> View App Logs).
• Adds alpha support for Ubuntu. I haven't actually tested this though, so please let me know if it doesn't work for you.
• Adds support for Thespis Mistral 7B v0.6.

Desktop:

• Adds Pygmalion 2 7B and Pygmalion 2 13B models.

Web & Desktop:

• Updates the create character page to include better suggestions on how to write character prompts that work with llama.cpp (Alpaca) models. Your old characters will likely still work, but you may want to edit them based on these new suggestions.
• Improves the experience when loading and unloading models.

Desktop:

• Updates to use llama.cpp.
• Adds several new LLM's - we now support: TheBloke/OpenHermes-2.5-Mistral-7B-GGUF, TheBloke/Luna-AI-Llama2-Uncensored-GGUF, TheBloke/Toppy-M-7B-GGUF, TheBloke/Nous-Hermes-Llama-2-7B-GGUF, TheBloke/Llama-2-7B-Chat-GGUF, and TheBloke/Llama-2-13B-chat-GGUF. Now that we're using GGUF it's easier to add new models, so feel free to request any others in ⁠🙏︱model-requests.
• Adds logs under Help -> View App Logs. We don't log anything about your chat here, just about model downloads and loading. If you find any bugs please include this as a text file in your report.

Web & Desktop:

• Fixes an issue with ordering of characters on the home page.
• Adds upvoting and downvoting to characters on the home page.
• Clear old messages when you reload the chat.

Desktop:

• Replaces node-ipc with @achrinza/node-ipc because of the protestware in node-ipc: https://www.lunasec.io/docs/blog/node-ipc-protestware/.
• Adds some computer info to the settings page.

Web & Desktop:

• Fix an issue with the model becoming incoherent if your chat exceeded the window size.
• Show the total number of pages on the characters page.
• Added links to back and forth between the Edit and Chat pages to make the chat <-> edit loop easier.
• Improve error handling during loading of the model.
• Add ability to configure the Temperature, Top P, and Repetition Penalty and persist these to local storage (i.e. a file on desktop and localStorage on web).

Desktop:

• Add a button which will show you where your config files and models are on your computer.