Interactive demo | lm/13-knowledge-base

def endpoint(mode:["question", "status", "deposit", "lookupWiki", "lookupWebsite"]="question", content:serve.text()="What is the panic monster?", category:str="", name:str="",
             model:["gpt-3.5-turbo", "text-davinci-003", "text-curie-001", "text-babbage-001", "text-ada-001"]="gpt-3.5-turbo", password:serve.text(password=True)="") -> serve.html():
    """Tool to quickly deposit articles and ask questions about all stored articles, powered by large language models.
There are several modes:

<h3>"question" mode</h3>

You can ask the system any question you'd like. It'll search the knowledge base for relevant
information and use that to answer your question in natural language. Active parameters:

<ul>
    <li>content: put your question here</li>
    <li>category: if specified, then searches in the specified category only, else searches in all categories</li>
    <li>model: the model that you want answer the question with</li>
</ul>

<h3>"status" mode</h3>

Gets general statistics on the knowledge base. No parameters are active.

<h3>"deposit" mode</h3>

Here, you can deposit articles into the system. It will take it in, splits up into multiple
sections, indexes and saves them to disk.

<ul>
    <li>content: put your article's contents here. Just copy-paste it in</li>
    <li>category: if specified, then dumps the article in a specific category, else dumps it in the default category</li>
    <li>name: if specified, then use this short name to identify the article, else it uses the current unix time</li>
    <li>password: the password necessary to deposit the information into</li>
</ul>

So, this isn't really free for the general public to deposit articles into. First and foremost,
it's a tool for myself. If you want a similar system, all the code is available for you to set
it up yourself, check out the source code <a href="/lm/13-knowledge-base">here</a>.

<h3>"lookupWiki" mode</h3>

In this mode, you can enter a website url in `content` parameter, like "RS-25 engine", and it will
automatically fire up chrome, look up on google for the term, grab all urls from wikipedia,
and finally index them.

<ul>
    <li>content: put your search term here</li>
    <li>password: the password necessary to look the term up</li>
</ul>

<h3>"lookupWebsite" mode</h3>

In this mode, you can enter a search term in `content` parameter, like "https://www.nature.com/articles/s41586-020-2528-x",
and it will automatically fire up chrome, go to the website, grab the article then index it.

<ul>
    <li>content: put your website url here</li>
    <li>password: the password necessary to look the term up</li>
</ul>"""
    try:
        if mode == "question": return question(content, category, model)
        if mode == "status": return status()
        if mode == "deposit": return deposit(content, category, name, password)
        if mode == "lookupWiki": return lookupWiki(content, password)
        if mode == "lookupWebsite": return lookupWebsite(content, password)
    except Exception as e:
        with k1.captureStdout(False) as out: traceback.print_exc()
        out = out() | join('\n')
        return f"<pre style='color: red'>{out}</pre>"