Folderr.com

Updated: 9/17/2024

Need help? Contact us at hello@folderr.com or join our Discord

Getting Started Guide

Table of Contents

How is Folderr different from ChatGPT?

AI models like ChatGPT’s OpenAI (4o, 4-turbo) are trained on vast amounts of data, but they don't always have access to real-time or specific information—like your personal documents or a company database. This is where RAG (Retrieval-Augmented Generation) comes in, which is a technique used at Folderr to improve responses.

Folderr allows you to connect your data with any LLM from Google Gemini, to GPT4o, even Claude and Llama. The Folderr app also allows you to connect your Assistants with tools - like internet search, image generation (Dalle3), and calculators.

If you’re not sure which model to use, stick with GPT 4o or 4o-mini. Both work great!

How do I get my AI Assistant to do what I want?

Instructions for the AI assistant should be configured and stored directly within the Assistant Settings page. For example:

If you want to create a sales support assistant trained on your products, you can provide specific instructions such as:

These custom instructions must be stored within the settings page, as they will not function effectively when uploaded in a document.

What is RAG?

RAG (Retrieval-Augmented Generation) is a system that enhances large language models (LLMs) by combining two steps:

  1. Retrieval: It searches a set of documents, files, or databases to find the most relevant information related to your question.
  2. Generation: Then, it feeds that retrieved information into the AI (LLM), which uses it to generate a more accurate and specific response.

This way, RAG helps the model give better answers by combining its general knowledge with specific, up-to-date, or personalized information.

How Does RAG Work Before Calling the LLM?

Before Folderr sends your question (or "prompt") to the language model, our RAG architecture does some extra work:

Why is RAG Useful?

Efficiency: RAG narrows down what the AI needs to look at, making the response faster and reducing the amount of unnecessary information.

Better answers: By pulling in specific, relevant details from a database, the AI can answer questions more accurately than if it were relying only on its general training.

Avoids context window limits: Language models like GPT-4 have a limit to how much text they can process at once. By filtering out unnecessary information through retrieval, RAG keeps the input size smaller and within the model’s capacity.

How can I import data Into Folderr?

Here are simple explanations for each data import option:

1. Import from Google Drive

This option allows you to bring files directly from your Google Drive account into Folderr. After connecting your Google account, you can select the specific files or folders you wish to import, like documents, PDFs, spreadsheets, or any other content stored on Google Drive.

2. Import from Dropbox

Similar to Google Drive, this option lets you connect your Dropbox account to import files or folders stored in Dropbox. You can choose specific files or entire directories to transfer into Folderr, making it easy to centralize your data.

3. Import from YouTube URL Transcript

With this option, you can import the transcript of any YouTube video by providing its URL. Folderr will automatically retrieve the text from the video’s transcript, allowing you to use or analyze that information within the platform.

4. Import from Website URL

You can import the content of a web page by entering its URL. Folderr will extract the text from the site, allowing you to store, organize, and use website content for further processing or analysis.

5. Import from a Sitemap

A sitemap is a file that lists all the pages on a website. By importing from a sitemap, Folderr can pull content from multiple pages of a site at once, helping you gather large amounts of data from a website efficiently.

6. Import from a JSON API

This option allows you to bring in data from a JSON API, which is a format used for exchanging data between systems. If a service provides an API in JSON format, you can use it to pull in structured data automatically, such as product listings, blog posts, or user data.

Note: This is a more complex import for organizations; please reach out if you need help importing via API.

7. Import from RSS

RSS feeds are often used to distribute blog posts, news articles, or other regularly updated content. With this option, you can import data from an RSS feed into Folderr, keeping your content up to date automatically.

Note: There are many RSS formats; if your RSS import doesn’t load in 1-2 minutes, please reach out to us at hello@folderr.com for help.

8. Import from Email

When you click on this, you’ll see the email address associated with your AI assistant. When you email the assistant URL, files will automatically upload to your assistant for training. Training does not happen automatically, and you will still need to click “TRAIN”.

9. Import from FTP

FTP (File Transfer Protocol) is a way to move files from one server to another. This option lets you import files from a remote FTP server into Folderr, making it easy to bring in large sets of files or data stored on external servers.

Training Structured Data (CSV and XLSX)

Folderr enhances the accuracy of prompt responses by uniquely analyzing your CSV and XLSX files. When you upload a file (CSV or XLSX), Folderr automatically converts it into a SQL table (a structured database table). Your prompt is then transformed into a precise SQL query, allowing Folderr to retrieve the exact data needed to provide accurate answers.

How It Works:

Note: CSV and XLSX files must be single-tab documents for this to work correctly.

In essence, Folderr uses SQL elements behind the scenes to turn your data into highly organized tables and deliver accurate, data-driven responses from your files.

Training Unstructured Data (PDF, Word, txt, etc)

Folderr processes unstructured data by converting it into vectors and storing them in a vector database. This allows for semantic search and better understanding of the content during interactions with your AI assistant.

What is a Vector Database?

A vector database stores data in the form of vectors—essentially lists of numbers that represent the meaning of text. In AI, when unstructured data (such as text) is processed, it's converted into vectors. These vectors capture the context, meaning, and relationships of the words, allowing the system to understand and compare them in a very sophisticated way.

Think of it like this: Instead of storing your data as plain text, a vector database holds it as complex numerical patterns that represent the "essence" of the content. This allows for much more powerful searching than traditional keyword searches.

Sharing your AI Assistants

Folderr allows assistants to be shared a few ways:

Public URL

Simply click "Copy Link," and you will have a URL for anyone to access and chat with your assistant. Users do not need a Folder account to chat with your public assistant.

Note: Tokens will be consumed from your (the Assistant owner’s) account.

Private Assistant

When you share an assistant via the “Edit Access” menu, you can insert their email, and then allow them to only chat with the assistant or to edit the assistant’s knowledge.

When sharing via email, the shared assistant will require the user to have a Folderr.com account, and tokens consumed will be consumed by the shared user—not the owner.

Restricted Access URL

You can also change the URL to be private, and only those with email access will have the ability to chat with your assistant.

Building an AI Chatbot for your Website

Folderr allows you to convert an AI assistant into a chatbot for your website.

Simply select the Assistant Name, provide it with an initial message, customize the display, and publish!

Domain Restriction

This is a security measure to prevent other domains from using your chat widget on their website.

List of domains where you want to embed your widget (e.g., example.com, subdomain.domain.net). Include only the domain and extension, without the protocol or any slashes (/).

Chatbot Lead Capture

Lead capture is a form asking the user for information before they are allowed to chat with your assistant.

Chatbot Code Snippet

Lastly, publish your chat widget and then click on the "EMBED" button to view and copy the generated code snippet.

Chatbot Insights

Each chat conversation will be recorded, and you will have the ability to download all of the conversations. You could even upload the conversation history back into your AI assistant to analyze the questions users are asking to better improve information on your site and better serve your customers and prospects.

Logs

Access detailed logs of all interactions to monitor performance and user engagement.

Chat Leads

When you have ‘Capture Leads’ active, all information requested will be stored on the leads page.

Workflows

Workflows are an incredibly powerful way to build automations. AI-powered workflows, aka ‘Agents’, connect different systems with tools and LLMs to perform complex tasks.

Folderr Workflows are still in BETA, and customers interested need to reach out to hello@folderr.com to get started.

Example 1: Automated GitHub Comments

Here is an example of how we use automation and AI to automatically collect changes we make to the Folderr.com code repository and have GPT4o Mini summarize the results in a way that can be understood by humans (not developers).

This automation reaches out to GitHub and posts a comment explaining all of our changes with zero human intervention.

Example 2: Financial Data Validation

Another example is a company that wanted to use Folderr as a deterministic rules engine to validate financial data.

They leverage our API to send files and questions, and get back results that can be analyzed and recorded in a third-party system:

October/2025

Example 3: RSS Automation

We’ve had requests for RSS automation, but an RSS is basically a list of websites. To automate this process we need to first collect all of the sites from the RSS URL, and then we have to use our ‘web scraping’ action node on every one of the URLs. Lastly, we want to send the collected data to your AI assistant and ‘retrain’ this process. This automation is not yet possible as we build out our looping capabilities, but is planned for October/2025.

Workflow Roadmap

Request DemoTry for free