maybe port/compile tiktoken to js (via tiktoken-rs to wasm, or tiktoken-py with pyodide)

Even more than talking to a character, I'm interested in how two characters talk to each other.

Character.ai has a feature where you can put multiple characters in a chat room (https://beta.character.ai/chats), but it still requires user input (even if it's just hitting enter to proceed). It'd be awesome to build up really well-prompted personas and just let them dive deep into conversation with each other.

when you add many starter characters no scroll down... i was trying to create many characters and adding to start but its not possible scrolldown only show : https://pasteboard.co/42v18BWKEwFY.png

I get an error message when I try to edit memories. I get the same error in both local host and on the public site. I don't know anything about coding, I just really enjoy this program for RP reasons. I have tried saving and uploading the character and text and restart everything but it doesn't help.

like this:

The frequency_penalty, presence_penalty, max_tokens are useful to tinker with the responses further, should be per character to allow maximum flexibility. Context limit would be how many tokens at most must be sent at max instead of using the limit of 4096, setting it at half helps to save on $$$ at the cost of some memory.

Maybe it's just me, and I'm lazy, but sometimes I'd just like to be able to press a button and watch how the conversation unfolds without me needing to do anything xD

I propose a feature that would allow AI to "roleplay" as me from each conversation.

There's already a button to delete messages, but it would be possible to add an button to insert messages too? Ideally, there would be a dropdown in the insert message popup, allowing to select the role for the new message (User / System / AI).

Hi. I'm searching for info related to the prompt that the character recive in relation to the memory generation (for example, how are the text splitted in chunks before embedding? Et simili... ) and lore recall. Thank you for your work!!!!

Currently, if you use /ai instruction to give the AI an instruction -- either on its own, or at the end of a normal user response -- the instruction is correctly reflected in the AI response. But if you then use the blue button to generate a variant of the response, the AI instruction is apparently not included in the new API call, and so is not reflected in the variant response.

(Thank you for this great project!)

I used the code to make my characters show ‘facial expressions’ or pictures but instead of showing the picture it just shows a black square with a blue question mark on my phone, and just the text for the emotion on my computer. I can’t ever see the picture. Would anyone happen to know what is going on here?

https://www.virustotal.com/gui/file/5692497e685b9c9f5a6866277a33ca37cbbdfa8771a26e1a33c97586f6a0d4b4?nocache=1

"JS:Downloader-AQL [Trj]" is a Trojan virus of the "downloader" type detected by antivirus software on a computer or device that operates on JavaScript (JS).

This type of virus typically infiltrates a device through a malicious website that can enter the computer when clicking a link or downloading a file. The downloader virus may download and execute other malicious programs on your device, which can behave unpredictably and affect the security of your data and personal information.

If your antivirus has detected "JS:Downloader-AQL [Trj]", it is recommended to immediately remove it using an antivirus program. It is also recommended to check the computer for other malware and to be more vigilant about internet security when using the device.

Sometimes we don't need to regenerate the whole message, but we just need it to be a bit longer.

I think it would be cool to allow users to generate another response from the AI just below it's previous answer.

Or maybe some kind of button "add more" attached to the AI's response.

edit: tested in safari on iOS

Like this tool! But I want to run it on my iPhone and I suspect the code is not being executed when using the Python coder.
I’ve also tried running in codespaces and seem to get similar results. Is there something I’m missing ?

This is the common answer:
‘’Sure! Here's an example Python code that creates a randomly initialized neural network with numpy, and prints the output given a random input:’’

it shows the code but no output or result.
Or if I ask: please run the Python code in the last message:

‘’ I apologize, but as an AI language model, I do not have access to the Python environment to run the code. However, I can assure you that the code I provided should work as expected. If you encounter any errors, please let me know and I'd be happy to help you troubleshoot them.’’

When I try to import a character data in TavernAI or Pygmalion JSON format, the following error appears.

It would be much easier if we could support the import of characters from other platforms(both JSON and Character Card image files).

You can refer to these two websites:

• AI Character Editor
• Character Hub

https://github.com/oobabooga/text-generation-webui/ can be used as a back-end to run dozens of different local models, including the latest LLaMA model. (LLaMA-13B beats GPT-3.5 in benchmarks while fitting in only 10GB of VRAM. LLaMA-30B goes well beyond GPT-3's capabilities while requiring only 20GB of VRAM to run locally.)

This error show when i send the api and try to use.

https://ibb.co/42WQ9TZ

There should be support for the curie model

It's great that we can setup characters with the system message. But as far as I know, there is currently no way to send system messages when you are already in an active conversation, right?

I think that the ability to send system messages anytime would work great as a "pointers" for AI into what direction take the conversation.

Once the conversation gets long enough to require summarization, it would be nice if there were a way to quickly tell what part of the start of the conversation is being summarized. For regular users, this could be a cue to mention important story elements every so often; for more sophisticated users, it could be a reminder to check the summary/memories/lore. A user can currently see this information by clicking the brain icon for the most recent AI response; but it could definitely be more elegant. A possibility: A "Summarization ends here" message displayed between the last summarized message and the first non-summarized one, displayed in the same way that the reminder message is displayed before the most recent the AI response.

Running tonight directly from https://josephrocca.github.io/OpenCharacters/#, I created a couple of new characters, and eventually had a very long chat with one (perhaps 200-300 message pairs? More? Hard to say). All was well, but suddenly I start getting the attached context_length_exceeded error. The length is exceeding 4097 by a very small number of tokens. The error repeats with each submission from this point on, although the token counts vary slightly. I tried deleting and resubmitting my last input, and many retries, but all now fail the same way. Switching characters to a new instance of the same AI works fine, then fails again when I switch back to the long chat. So my long chat is basically over now as a result.

The last input that first triggered this is nothing special, and is in fact rather short. So I think there's something off in your max context length calculation. I saved the thread at that point and can provide that and/or full screenshot or bot name if required.

I noticed that the web app seems to take longer and longer to summarize with every new prompt that i make,i figure this is normal but then at the last prompt it gives me this error message which i included in the image,

I know this uses GPT trough OpenAi's API but it should be very useful if integrating it with some GPT-J API or other models so people can run they own model servers and use locally the app?

(i.e. allow the thread model to override the character model)

Title.

Ask for it to do anything in HTML, the thing shuffles in madness. Will you fix it so the output and input is properly escaped?

Thanks

Add an image URL "type" to the prompt2 utility - it displays the image to the right of the URL

Even if you change to drop older it always comes back to summarize.

I'm looking into what it'd take to write an exporter for Notebook.ai that lets users either import individual characters or all of their characters into OpenCharacters. Being able to talk with the characters you've been building seems like an awesome feature for authors and other creatives who want to make their characters feel a little more real before, during, or after writing stories about them.

A lot of this is just thinking out loud and translating my interpretation of your code to words, but there are a few questions below too. :)

I'm using Zoltanai's character editor's JSON format as a template for a valid import format, which includes the following fields:

• char_name
• char_persona
• world_scenario
• char_greeting
• example_dialogue
• name
• description
• personality
• scenario
• first_mes
• mes_example

Looking at the exported values from the tool, it looks like a few of these fields are redundant to cover importing to different tools expecting various character formats, so I just want to make sure I'm reducing the set to what's actually used here and packing as much information into those keys.

Looking at tryImportingExternalCharacterFileFormat, it looks like the generic mapping is as follows:

[1] I don't see anything in the import code to set reminderMessage, but I do see it gets used elsewhere (and can be set from the UI). Is this something available for imports and I just missed where it's set?

It looks like just name, avatar, personality, description, scenario, example_dialogue, and char_greeting are the used keys for a full import. Am I missing anything?

I'll try to get a generic export prototyped in Notebook.ai soon so I can see how well it works OOTB with the current generic import functionality, but if there are other details that would be helpful to each chatbot to have and improve quality on your end, I'd definitely be open to exporting those too! :)

A couple extra questions (or maybe feature requests) on the import process also:

1. Does OpenCharacters support importing multiple characters at once, or is it always just one at a time? It'd be great to let users easily import all of their characters at once and just pick and choose which one to chat with from the standard "new chat" UI.
2. Is there an endpoint I can link to and code the export data to have it automatically import / populate character(s)? For example, it'd be great to have a "Talk to this character" link on my end that routes a user to OpenCharacters with that character's information and lets them immediately start chatting, as opposed to offering an export download to be re-uploaded as an import and then start chatting.

I have been experiencing trouble while trying to use DaVinci. GPT works just fine, but whenever I switch to DaVinci, the completion shows up as incomplete, cut off. I'm suspecting it has something to do with max_tokens. I think adding more ways to tinker with the character, such as max_tokens, top P, top K or rep penalty would be quite good. What should I do to fix this in the meantime?

I had this issue a while back and it's happening again. I changed a bot's model from GPT 4 to 3.5 and regenerated it's message. It wasn't generating anything, so I refreshed the page. Upon refreshing, suddenly the chats are gone/unable to render. Cleaned my cache and cookies but it doesn't work. Any help?

You recently enhanced OpenCharacters so that AI instructions (given using the /ai command) are saved in the conversation history.

Now that that's done, it would be nice if the AI instruction was included in the 'Here's some data that the character used to generate this message:' dialog that appears when you click on the brain icon next to an AI response.

Thanks again for the extremely useful project!

AI Horde is a crowdsourced cluster of workers for text and image generation that can be used anonymously and/or free. It provides a fully documented REST API you can easily integrate to your front-end. You can also tie it into generating images for further immersion in the chat.

I am the creator, so feel free to hit me up on questions on discord.

I get this javascript alert after every message (as well as after refreshing the page on any chat where a message is loading):

In the console, I also see this stack trace:

(index):3324 TypeError: Cannot read properties of undefined (reading 'digest')
at sha256Text (utils.js?v=23:342:42)
at countTokensInMessages ((index):1972:28)
at getBotReply ((index):2509:40)
at async doBotReplyIfNeeded ((index):3296:67)

The exact line throwing the error is

const hashBuffer = await crypto.subtle.digest('SHA-256', msgUint8);

I'm not sure if this is perhaps just a setup issue on my part, so I've also included my setup instructions below.

Setup details

I'm running the project zip from the local setup doc locally and set my OpenAI key. I'm also serving the project with Apache instead of the Web Server for Chrome app, since I'm planning to host on a remote, headless server.

I'm pretty sure, according to this documentation on Crypto: subtle that the problem is that I'm not serving the page over HTTPS and therefore crypto.subtle is unavailable.

Just figured I'd open this issue (and hopefully close it shortly) while I dig in for myself and anyone else that sees this specific error in the future. 👍

• completely manual summaries?
• inform LLM which facts must be included in the summary
• ?

Changing the system message of the character would allow changing the personality of the character over time or adding context without putting it in a new user message as that could cause the ai to get confused between actual user message and the contextual message. Examples would be updating summary, inventory, relationships, events or even relevant past conversations using sematic search etc for the character only for that thread in the system message giving the character a "long-term memory".

This could be done using the processMessages function allowing to add to/overwrite the existing system prompt on a per thread basis so that other threads of same character are not affected.

Edit: I just saw the api.chat and api.completion in the custom-code.md. Would this allow to replace the assistant message on it's turn using the one generated using api?

I mostly use OpenCharacters from my phone, on the default Chrome browser. Everything was fine, I went to have lunch, and when I came back to keep chatting, I got an error message (can't really remember what it said). I refreshed the tab and now everything is gone. Is there any way to recover what I lost? This is extremely frustrating.

You can make GPT remember more data by taking advantage of longchain. In addition, active consciousness is limited to 40 instantaneous processes, while passive consciousness is billions of neurons. Maybe instead of trying to free GPT, we should let it think for itself and roam freely.

• recency bias
• explicit triggers that are "outside" of a memory
• LLM-based memory ranking (i.e. ask GPT which memories are important and use that in weighting)
• (?) look at sota papers for more ideas

i was trying to set by default the ( let openAiApiKeyOriginal = (await db.misc.get("openAiApiKey"))?.value || "sk---------"; )

but then can not hide the apikey... and every time force the label promp:

let result = await prompt2({
openAiApiKey: { label: "Please create a new OpenAI API key and paste it here. Go to this page to do that. You can change or delete this later by clicking the 'settings' button.", type:"textLine" },
});

im trying to hide the default api key
openAiApiKey: {label: "OpenAI API key:", type: "textLine", placeholder: "(optional)",defaultValue: openAiApiKeyOriginal, readOnly: true, hidden: true },

but not works
....woulbe interesting when you set the default apikey value do not force que promp.

its only an suggestion.! Great job! :D

josephrocca.github.io says
Please report this error on the Discord or Github:

stack: DatabaseClosedError: UnknownError Internal error opening
backing store for indexedDB.open.
UnknownError: Internal error opening backing store for
indexedDB.open.
at t (https://josephrocca.github.io/OpenCharacters/:10:18817)
at https://josephrocca.github.io/OpenCharacters/:10:19017
at https://josephrocca.github.io/OpenCharacters/:10:16578
at Fe (https://josephrocca.github.io/OpenCharacters/:10:10996)
at Ue (https:/bosephrocca.github.io/OpenCharacters/:10:11472)
at Ne (https:/josephrocca.github.io/OpenCharacters/:10:11326)
line: 0

Inputting a long response that does not go into a new line causes the page to increase in dimensions, specifically at the elongated response.

How to replicate

Input a response like aaaaaaaaaaaaaaaaaaaaaaaaaaaaaa You get what I mean

I have chatGPT open in one window and this in another both using 3.5. The responses take minutes to return vs seconds via OpenAI app. I get that there is an interpreter involved, a middle man, but what can I do to speed it up, what is causing the delay, can I fix it?

i read that phrases in docs : Create your own memory structures (embedding, retrieval, etc.)

Improve your character's memory by setting up your own embedding/retrieval system

is this already implemented?

How can I enable character storage?

From what it reads, all you need to do change the endpoint: https://goose.ai/docs/api to transition from OpenAI to GooseAI. It's a lot cheaper and as far as I can see, doesn't need "jailbroken". It would be amazing if this was added.

Mainly need to work out where this stuff should go in the UI

First of all, thank you for this awesome project and the constant iterations of updates.

I found that when I imported the TavernAI format, the personality section was missing information. In the TavernAI format, it is in the form of a description.

I found the following code which can be used as a reference to differentiate between the various char formats.
Constructor
https://github.com/ZoltanAI/character-editor/blob/main/index.html#L1321
Exporter
https://github.com/ZoltanAI/character-editor/blob/main/index.html#L1745

For example, from the section in the image below, we can see an example of how it is imported as the TavernAI format.

Export to (Pygmalion / Text Generation JSON)