The Basics

Q1: What is this project?

Holo AI is a writing aid that allows you to use advanced AI to guide and advance your content. It's intended to support both regular writing tasks and its use as a roleplaying game. This faq was written with the help of Holo AI, to fix my writing block in making a public announcement about these questions.

Q2: How does Holo AI help me write things?

Anytime during the course of your writing or exploring, you can click the "generate completions" button. The AI will give you several options to continue whatever you've written on its own. This is unlike any "autocomplete" you are probably be familiar with. Given the right prompt, the AI is capable of generating entirely new and coherent content.

Q3: What are all these fancy settings? Memory, world info...

Most of the settings fall into two categories: Those that tune the AI's output, and those that affect the AI's "field-of-view". The "output" settings, mostly under Datasets, are ways to inform the AI about what in general you're writing. For instance, you can 'tag' a document as fantasy, and the AI will continue the document with the understanding that it's got such a tag. The "context" or "field-of-view" settings determine what the AI is prompted with. Currently, large language models like ours can only see a number of words back, in this case around 1800~. In order to prevent the AI from forgetting about essential parts of your story or document, you can use the world info and memory settings to keep them in view. For a more extensive guide rather than an overview, you can read a guide by one of our users here.

About Us

Q4: Who are you?

I, the person writing this, am Dean, aka lc. Holo AI's development is done by Etienne (tma) and I under Leonard Cyber, a cybersecurity company that, among other things, performs aptitude testing for application security pentesters.

Q5: What are application security engineers doing making Holo AI?

Since we're a two person company and can do whatever we want, we think of ourselves as solid developers, and we thought the project seemed exciting, we released a demo. The response greatly surpassed our expectations, so now we're doing our best working on it full time.

Security and Privacy

Q6: How do I know my data is secure?

Leonard Cyber is a development team solely comprised of application security engineers and pentesters. The frameworks we've chosen to build our application don't come with the wiggleroom that allows us to easily commit common security mistakes like SQLi, XSS, CSRF, etc. They're also not the latest and greatest JS overhauls complete with ten different template injection and prototype pollution bugs lying around. That being said, just because an app was designed by information security professionals doesn't mean that it's secure. Holo AI is also being updated multiple times a day, and is still well in beta. If you are trying to find out where to draft weekly intelligence updates to the FSB, you probably shouldn't trust web applications written by random internet people who say they're security experts.

Q7: Do you do client-side encryption of user stories?

Short answer: Yes. You must however register an account first, because your stories can't be encrypted if you don't have a password yet. Once you register your account, every story you create will be encrypted with a key derived from your password, which never gets sent to our servers thanks to a cool key exchange technique called Secure Remote Passwords. In around a week the app will begin to retroactively encrypt stories as you access them. We personally use end to end encrypted services for things like email, SMS, and file storage and think they're great. However, here there's always going to be a lingering problem: It's impossible to run a language model this size within your browser, so in order to use the application, you have to send us the plaintext prompts. This is a design flaw with an end-to-end encrypted AIDungeon service that can't be overcome unless you're willing to rent your own dedicated enterprise GPUs required to run the language models yourself.

Q9: Terms of Service?

Terms of Service

Start writing!