GPT-5, GPT-5-mini, Gemini 2.5 Flash-lite / Flash / Pro, Grok 4, Claude Sonnet 4.0, Perplexity, Kimi K2 1T, DeepSeek R-1, Qwen3-32b, Llama 4 Maverick. New models are added as they get released. The Kimi K2 1T, Deepseek, and Qwen Chinese models are served from an isolated stateside server, so no data is ever shared nor touches a Chinese API.

What is RAG?

RAG is a technique that augments LLM knowledge with data not present in its pretraining dataset. While an LLM's knowledge is "frozen" at its training cutoff date, RAG enables it to access and utilize your personal or proprietary data.

How RAG Works

RAG converts your documents into tokens and vector embeddings—essentially translating them into a language AI can understand. This is similar to what happens when you upload a file directly to an LLM, but with significant advantages.

When You Need a RAG Platform

A dedicated RAG platform becomes necessary when:

1. Dealing with documents too large for the AI model's context window
2. Working with many documents simultaneously
3. Requiring long-term AI memory of all your information
4. Using file formats not natively supported by LLMs (Google Docs, MS Office, etc.)
5. Wanting to share your knowledge as an AI interface others can query

The Power of RAG

What makes RAG powerful is its precision in:

• Using AI to find relevant content chunks across all vectorized documents
• Feeding only the most pertinent information to the LLM as context
• Performing "similarity searches" by converting queries into vector embeddings
• Identifying vectors closest to the query in a high-dimensional mathematical space

The Result

This approach enables highly targeted context retrieval for any query, even from massive document collections. The more specific and well-formulated the question, the better the system can map it to the relevant "nearest neighbor" vectors containing the answer.

AskMyAI utilizes the highest security standards. SOC 2 Type II compliant; HIPAA compliant; GDPR compliant

• End-to-End Encryption: Any data in motion is encrypted end-to-end
• Google Cloud Storage: For data at rest, volumes are stored on Google Cloud Storage using Google's security protocols
• Signed URLs: When you (or someone you share a collection with) queries a collection, the source text is provided by AI is a signed URL with policy expiration of 1 hour.
• Data Privacy: We do not sell your data or use your data for any purpose other than providing the AskMyAI service. We do not train our own models on your data.
• Access Control: If you share a collection with other AskMyAI users to appear in their drop-down of collections, you can adjust their permission to Editor or Viewer. Even if editor, they cannot directly modify or delete any source document you vectorized to your collection. They can only modify/delete their own documents they added to your collection.

AskMyAI has special handling for the broadest range of file format support from a RAG platform in the marketplace. This includes, but not limited to, Google files (Docs, Sheets, Slides), MS Office files (.docx, .xlsx, .pptx), images, video files, audio files, CSV, multi-tab XLSX files, PDF, markdown, and even Kindle e-books

No

No, but we recommend limiting the number of docs to a single collection to 100 for optimal retrieval performance. If it is a very large document with hundreds of pages, we would recommend putting it on its own dedicated collection.

Yes. While nearly all URLs can be scraped and vectorized in a single-click on the AskMyAI platform, not all URLs and social media posts can be scraped. If the social media post is set to "private" rather than "public", it will not be scraped. If the website is gated with a login form or uses certain protections against programmatic access, it will not be scraped.

NO. Agentic safeguards are put in place to ensure no hallucinations occur. If AI cannot find the answer in the provided context from your collection, rather than hallucinate an answer, it will respond with "The answer is not contained in the provided context."

First step is a scoping session with our services team to understand the need. Key considerations:

Data Access & Processing

• Source method: Is the AskMyAI platform fetching data directly from your system(s) via API, or are you sending it via webhook, SFTP, an AWS/GCS volume, etc.?
• Query complexity: If AskMyAI is making the fetch, is the database query hard-coded? Is the schema simple enough that an AI agent can be built to compile the backend query from unstructured text from a user? Or is the schema and conditional parameters complex where we need to customize our UI to use the parameters, conditions and filters your users need to compile the right query?
  • For this reason, we have an easily-customizable Streamlit interface that is integrated with our AI interface, ETL processing, and vector database to handle the data transformation and storage
• Triggers: Is this an automatic job that runs on triggers or manually triggered by the user? If triggers, what are we listening for (webhook, email, cron job, etc.)
• Data processing: Define the "T" in "ETL" on how fetched data needs to be transformed to be AI-friendly prior to embeddings

Yes. Once you have vectorized your knowledge to a collection and integrated your systems to fetch real-time data if needed, you're ready to automate workflows using the Agent Builder tool.

The Agent Builder makes it easy to create custom multi-agent workflows that automate tasks.

Here's how it works:

1. Create Your First Agent:

   • Drag and drop the necessary elements onto your canvas to define the tasks your first agent will perform.

2. Empower Your Agent:

   • Knowledge: Leverage information stored in your collections.
   • Data: Integrate real-time business data from your connected systems.
   • Instructions: Provide specific prompts or questions to guide the agent.
   • Tools: Utilize functions like web search, youtube transcription, web scraping to extract content from a URL, etc.

3. Chain Agents Together:

   • Build powerful workflows by connecting multiple agents, allowing the output of one agent to serve as the input for another.

Example Workflow:

• Agent 1: Retrieves and applies knowledge relevant to the problem.
• Agent 2: Transforms data and uses tools if needed.
• Agent 3: Analyzes the outcome to take the appropriate action.

With "Agent Builder", you can rapidly prototype, validate, and deploy automated workflows tailored specifically to your business needs, without the need for any coding.

The good news is that by taking that step, you already understand RAG in its simplest form because that IS RAG. You are feeding the model your data as the context to answer the question. This approach works perfectly for small documents that fit inside a model's context window AND for a document with information you never want to retrieve again. There are 4 main reasons when a RAG approach is your only option:

• REASON 1 -> Large Doc or Many Docs If the document is large (as in longer than a typical chapter) or the answer is contained somewhere across hundreds of docs, you need a RAG system that can match your query to the most relevant parts of a document that are applicable as context to answer the query. This way the AI model is provided only the most focused content as context to answer the query that can fit inside its context window.
• REASON 2 -> Long-term Memory If you want data or information to persist as knowledge that is never forgotten by your AI and instantly available to be retrieved by your AI as context for any applicable query. For example, if you watch a 2-hour long youtube video and a subject comes up that you want to save for future reference, you can use ETL + RAG to transcribe the text, then vectorize it to a collection in your vector database. If you want to retrieve that information years later, RAG will retrieve the most relevant parts from that video as context for the AI model to answer your query. The source material is cited in the answer so you can click the URL to re-watch or even click the source file link to the plain text transcript.
• REASON 3 -> Share Knowledge If you want to share knowledge that is easily retrievable via AI. In the past you could only share knowledge as raw documents or videos. With RAG, you can share an AI interface that has access to a collection(s) in your "AI library" that you want to share. Users can make any query of the knowledge in your collection to get the precise answer they need along with the link to the full source should they want to access the entire file.
• REASON 4 -> Data in Various Formats LLMs are HIGHLY limited on the number of file formats supported. They also have limited URL scraping capabilities. One of the key functions of an automated ETL + RAG platform is the heavy-lifting on the pre-processing required before the embeddings step that can ingest documents from any format or source (e.g. URL, video, social media post) and handle the required scraping/parsing/transcribing to extract the text then convert into a markdown format AI can work with that retains the structure and hierarchy of the original non-markdown document, then add the necessary metadata to assist the model with context continuity during retrieval and citing of sources

Same Models, Different Knowledge Sources

They are the same models but use different data sets for its knowledge. The "regular" LLM like ChatGPT that you use directly on the OpenAI website or app is using public internet data it has been trained on for its knowledge base. The AskMyAI platform uses the data you vectorized to your "AI library" on the AskMyAI platform to augment the knowledge it has with YOUR data. What is also different is that the LLM MUST cite the source it used to find the answer from your AI library.

What Makes this Unique

• This is what makes it your own personal LLM that is an expert on anything and everything you want it to know

• When you ask the LLM a question about any data you vectorized to your "AI library" on the AskMyAI platform, the LLM does not attempt to fetch the answer from its generalize pretraining data. Instead, it is fetching the answer from the source material (documents, videos, URLs, etc.) you vectorized to a collection in your "AI library". If the answer does not exist in your "AI library", it will say it cannot find the answer instead of hallucinating an answer

• The real power of the AskMyAI platform is in the vector embeddings, vector db, and retrieval techniques that can fetch only the relevant parts of any document to the LLM as context to answer the query. This is the "magic" of a RAG system - to retrieve only the relevant context AI needs to respond to your query, even if that information is buried across hundreds to thousands of documents in a collection.

Platform Capabilities

The AskMyAI platform enables you to:

• Curate your "AI library" from any source in almost any format across hundreds to thousands of documents, videos, audio files, web urls, social media posts, images, kindle ebooks, etc. that you (or anyone else) can then use a leading LLM to query that data and performs jobs

• When AI answers your query, links are provided to the source material where you can seen the plain text file itself that was parsed from the original format or URL if you sourced the information from the web or social

• The data you vectorize persists as long-term memory for AI until you decide to purge. Your "AI library" essentially becomes a memory hack whereby you can retrieve anything that was important enough to save. This means one-click ability to vectorize any document, article, video, social post, or thought that you DO NOT WANT TO FORGET (or want to share). You may not be able to remember the specifics discussed in a lengthy youtube video, but you will remember the topic. And by asking the associated topic-related question, the AskMyAI platform will fetch the most relevant parts of any content you vectorized as context.

• Nearly any URL can be scraped

  • This means URLs from websites, youtube videos, social media posts, articles from the Apple News app and more

• Nearly any file can be parsed then vectorized to a collection with a single-click or automatically with the Google Drive / Microsoft Onedrive integration. See above for list of supported file formats

• Any thought can be spoken to the interface (or typed)

  • ...so the AskMyAI platform can ingest and vectorize to a collection that will always be available to you or anyone you share it with.

Yes.

While most knowledge is locked up on our head, hard drive, notebook, file cabinet, sharepoint and not easily shared ... and most often forgotten, AskMyAI allows you to share a collection from your "AI library" with anyone so they can use the latest AI models to retrieve relevant information from your collection(s) and respond to any query they have about data in your collection. Collections can be shared 2-ways:

1. Collections can be shared with other AskMyAI users to appear in their drop-down menu of collections to choose from.
   • Choose permissions like "Editor" or "Viewer"
2. Collections can be shared to a dedicated URL (hosted by AskMyAI for $5/month) so anyone with the link can chat with AI about the data in your collection without having to sign-up or register for an AskMyAI account
   • If sharing a collection to a dedicated URL, there is no limit on the number of collections that can be shared to the URL. Users with the link simply begin by choosing the collection they want to query from the drop-down menu
   • For privacy, an access code can be applied in which case, a user must have both the URL and the access code.
   • For stricter privacy, only share collections with other registered AskMyAI users
3. Collections can be shared as a chatbot to your website by inserting 5 lines of code

No. AskMyAI is browser-based. It is both mobile and desktop friendly. While the desktop experience is bullet-proof, we notice the mobile experience can frustratingly error at times which means refreshing the page to try again. When you do, it is always successful, but it is frustrating to have to refresh the page. For this reason we are rebuilding the front end to match the framework we used for the management console

Limitations of Common RAG Systems

Retrieval Performance Issues

• Poor semantic retrieval performance due to:
  • Lack of contextual pre-processing prior to embeddings that results in massive holes of missing contextual awareness across vectors
  • Inability to control retrieval size with any query
  • Inability to narrow the scope of the semantic search with metadata filtering for:
    • Specific document in a collection
    • Doc types
    • Date range
    • Keyword/phrase
    • User who vectorized the doc

Processing Limitations

• Lack of post-processing with retrieved vectors to re-rank, QA relevancy, and properly stitch contextual awareness for cohesive context prior to LLM injection
• Limited support of LLM models

Collection and Document Constraints

• Limited number of vector db collections allowed and/or limited number of documents per collection
• Highly limited file formats supported
• Limited document sizes

Integration Challenges

• Little to no support for auto-syncing hundreds to thousands of files in Google Drive folders or MS OneDrive folders
• Little to no web scraping capabilities across web and social
• Little to no support to transcribe large audio and video files then vectorize
• Little to no systems/database integrations to vectorize snapshots of real-time data to a collection

Data Structure Support

• Little to no support for tabular data structure vectorization & retrieval from a vector db
  • Even IF tabular structure support for CSV, zero support for multi-tab XLSX files
• Little to no retaining of document or web page hierarchical structure when parsed prior to vector embeddings

File Management Deficiencies

• Near-zero file management capabilities
  • E.g. edit source files, copy/move docs to another collection, rename vector db collections, sorting, auto-detect replacement of existing doc with an updated doc to avoid content duplication in collections (as long as the filename and extension remains the same), inline doc summarization

Metadata and Sharing Limitations

• Limited custom metadata capabilities
  • Zero support for adding additional context to any document upload or web/social URL scrape prior to embeddings to assist AI on retrieval
• Limited to no sharing capabilities
  • No permission granting abilities

Cost and Scalability Issues

• Expensive with no scaling abilities or storage tiering

Voice and Speech Features

• No speech capabilities to transcribe voice and vectorize to a collection
• No speech capabilities to query a collection through a browser

Workflow Automation

• No "Actions" capabilities to leverage the vectorized data in collections with a custom agentic workflow to automate tasks
  • The AskMyAI services teams works with organizations vectorizing content to collections on how to best leverage their growing knowledge base with agentic workflows that can do more than answer questions about data but use it to automate tasks when YOUR systems are integrated with the AskMyAI platform to put your collections to work

The Secret Sauce of High-Performance RAG

Pre-processing and post-processing are the secret sauce of a high-performing RAG system.

Pre-processing Elements

Before we vectorize your data, we clean it up and structure it in a way that maximizes retrieval accuracy. This involves things like:

• Context Continuity: The process of breaking down documents into chunks that become their own vector embedding can leave large context continuity holes without sophisticated techniques and agentic workflows to ensure these chunks maintain context continuity across vectors. This also means special handling for various data sources (e.g. chunking a video transcript is different from chunking a spreadsheet)

• Structure Retention: We preserve the original document's structure (headings, lists, tables) in a format the LLM can understand (Markdown). This is crucial for retrieval understanding the relationships between different parts of a document.

• Metadata Addition: We add metadata (like the source URL, document type, date, etc.) to each chunk. This allows for powerful filtering during retrieval.

Post-processing Techniques

After we retrieve the most relevant chunks, we further refine them before sending them to the LLM. This includes:

• Re-ranking: We use additional algorithms to re-order the retrieved chunks based on relevance.

• Vector Overlap Removal: We eliminate redundant information between chunks.

• Chunk Stitching: We intelligently combine related chunks to provide cohesive context for the LLM.

• Contextual Irrelevance QA: We use agentic workflows to identify and discard any retrieved chunks that, despite seeming relevant based on vector similarity, are actually irrelevant to the user's query.

• Metadata Injections: We can include relevant metadata in the context provided to the LLM, further improving its understanding.

Why It Matters

Without proper pre- and post-processing, RAG systems often return irrelevant or incomplete results. Our expertise in these areas is what makes AskMyAI's retrieval so accurate and reliable.

AskMyAI uses specialized handling for tabular data, recognizing that it's often structured differently than free-form text.

• Structure Preservation: We retain the table structure during vectorization, so the LLM understands the relationships between rows and columns.
• Multi-Tab Support: We handle multi-tab XLSX files, preserving the structure of each tab. This is crucial for complex spreadsheets where data is organized across multiple sheets.
• Querying Tables: You can ask questions that require the LLM to analyze and synthesize information from tables, just like you would with any other data type.

This is a unique feature of AskMyAI, designed to allow you to include your legally purchased Kindle books in your "AI Library."

• How it Works: We've developed a process to remove the DRM (Digital Rights Management) protection from your Kindle e-books. This allows us to extract the plain text, which we then vectorize.
• Validation: To ensure this feature is only used for books you own, we require you to provide the serial number of your Kindle device (in its settings). We match this against your ebook purchase to verify ownership.
• Legality: It's important to understand that distributing copyrighted material is illegal. However, removing DRM for personal use, particularly for purposes like this (making your own content searchable), generally falls under fair use principles. We do not store the plain text of your e-books; only the vectorized representation remains in your "AI library."
• Future-Proofing: While our DRM removal process works with the current generation of Kindle e-books, we cannot guarantee it will continue to work with future updates.

Follow these steps to vectorize your Kindle eBook:

1. Plug your Kindle into your computer via USB.
2. Navigate to \documents\Downloads\Items01
3. Locate the KFX file (example-kindle-filename.kfx) and folder sharing the same book title.
4. Copy both the .kfx file and folder to your computer.
5. Create a .zip file containing both the .kfx file and folder.
6. Upload the .zip file to AskMyAI.

This integration makes it easy to keep files (or even entire folders) in your Google Drive or MS OneDrive auto-synced with collections on the AskMyAI platform so any change to a document is auto-synced every 24-hrs.

• Batch Upload: You can connect a specific files or entire folders from your Google Drive or Microsoft OneDrive to an AskMyAI collection. AskMyAI automatically processes each file and vectorizes all supported files within that folder.
• Auto-Sync: While users can click the "sync" button on a file or folder to manually sync with the AskMyAI collection, AskMyAI monitors the synced files or folders for changes then automatically updates your AskMyAI collection every 24 hours. This ensures your AI always has the latest information and you never have to manually upload a document to it.

The link to the original source material is supplied with every AI response to show where it sourced its answer from. If the data came from a web URL you used AskMyAI to scrape and vectorize, the original URL will be supplied.

• If the originating data came from a file in your collection, you the link to the plain markdown text parsed from the original file will be supplied.

Yes! AskMyAI now supports Single Sign-On through Google and Microsoft accounts via Firebase. You can sign in using your Google Workspace or Microsoft Azure AD credentials. Additional SSO providers (e.g. Okta/SAML) will be added in the future - contact support@askmyai.ai for specific provider requests.

Yes! You can embed a collection from your "AI Library" as a chatbot on your website that is fully hosted by AskMyAI. This allows your website visitors to interact with your data using natural language. To do this, you'll need to insert a small snippet of code from AskMyAI's chatbot builder page into your website's HTML.

There is no limit on queries you make in the AskMyAI chat interface. Heavy usage of the latest state-of-the-art LLMs considered "Premium" models could mean rate limiting with access restricted to standard models like Gemini 2.5 Flash and similar class until the 1st of the next month. To continue using premium models, you can add a Pay-as-you-go balance to your account for continued access to premium models without restriction.

While queries you make in the AskMyAI chat interface are covered in the 'Core' plan, access to premium features are offered on a pay-as-you go basis.

Here is the breakout on premium features:

• Collections Shared to a Dedicated URL or Website Chatbot: Queries that OTHERS make to collections you share to a dedicated URL (hosted by AskMyAI) or as a chatbot on your website are charged a flat fee rate of $0.005 per query.
• Agent Builder: Any queries you make through the Agent Builder tool to run agentic workflows that automate tasks are charged at the market rate listed on the model rate card page which is merely a pass-thru charge
• Booking Page: Queries that users make on your dedicated booking page to schedule an appointment on your calendar are not charged by the query. The usage charge is $0.05 per booked appointment, which includes confirmation emails and follow-up reminders where the user can confirm/cancel/reschedule. Optional SMS reminders are extra at $0.07 per reminder

$5 in Pay-as-you-go usage is free to get started. After that, you must add a Pay-as-you-go balance to your account to cover usage for these premium features.

• Visit the Usage page for full transparency on pay-as-you-go usage costs for queries made against collections you share publicly or through the Agent Builder tool to automate tasks
• Example: If you are an Airbnb host, you can query your own collections directly through the AskMyAI chat interface as often as you like. If you share a collection to a dedicated URL that is a QR code on the fridge for guests to access, you will have to add a balance to your Pay-as-you-go balance to cover that usage at the rate of $0.005 per query. If you create agentic workflows using the Agent Builder tool to automate biz tasks, that usage will be deducted from your Pay-as-you-go balance as a pass-thru charge.

• 2,000 documents are included in the Core 2K plan
• 5,000 documents are included in the Core 5K plan
• 10,000 documents are included in the Core 10K plan
• For 10K+ documents, inquiry for enterprise pricing @ contact-us. For the latest pricing, go to the Billing page.

A document is a single file, URL, or voice recording. A single file can be as large as a book like War & Peace. It is also as small as the content from a single URL that is scraped for text then added to a collection in your vector database on the AskMyAI platform.

The hosting fee for a dedicated URL or chatbot is $5 per month. Anyone with the URL will be able to access without having to register for an AskMyAI account. You can create multiple dedicated URLs for hosting different collections, multiple chatbots for different web pages and web projects, and multiple booking pages.

There is a 25MB file limit for video/audio files as part of a Google Drive or MS OneDrive sync. If uploading video/audio directly in the AskMyAI chat interface from your local drive, the max size restriction is increased to 1GB that need to be transcribed then vectorized to your collection.

AskMyAI uses agentic workflows to analyze images and call the appropriate OCR engine for the job. This allows for extraction of text from images, handwritten notes, scanned documents, receipts, and other visual content, which is then vectorized into your collections for easy retrieval.

Yes. Here is how it works:

1. Integrate one or multiple calendars from various providers like Google, iCloud, MS Outlook on the AskMyAI "Calendar" page under the "Actions" tab
2. Sync one or more calendars to a collection
   • For individuals, if you choose to connect multiple calendars (work and personal) to a single collection, events from all those calendars will be combined so AI can see your full schedule across many calendars
   • For businesses, each calendar can represent a different employee or department. Choose to connect each calendar to its own collection OR connect all calendars to a single collection if you want AI to see what is happening across your entire company
3. Go to the AskMyAI interface and choose the calendar collection to start chatting with your schedule. It will fetch your real-time calendar data each time you do this.
4. [Optional] Share your calendar collection with other AskMyAI users or create a dedicated URL that anyone can use to talk to your schedule. a. Go to the Collections page to set this up after syncing your calendar(s) to a collection.
5. [Optional] After syncing your calendar(s) to a collection, go to the AskMyAI "Actions" page to create a dedicated booking page so anyone can book time on your calendar and receive reminders (text and/or email).

The Conversion Process

"Vectorizing" is the process of taking plain text parsed from a document and turning it into a numerical representation that AI can understand.

The Coordinate Metaphor

Think of it like this: every word, phrase, and concept has a unique "coordinate" in a multi-dimensional space. Words with similar meanings have coordinates that are close together. The same is true for ideas, concepts, and topics.

Creating an Information Map

When you vectorize your documents, we are using OpenAI's embedding tool to create a map of all the information contained in your document(s). When you ask a question, we convert that question into vector coordinates as well then find the closest matching vector coordinates from your source material that was vectorized to a collection in your "AI library" that represents the most relevant text associated to the question. In technical terms, this known as semantic search. In layman's terms, this is known as a "nearest neighbor search".

Semantic Search in Action

This is how semantic search works. Critically, this means that only the most relevant text from any document or series of documents is retrieved as context for the LLM to answer your question even if the question is queries against a collection containing 1000s of documents.