Why do I need a knowledge base?
You have hundreds of files and want to work with the knowledge inside all of them at the same time? That's what knowledge bases are for. You can seamlessly retrieve information from all your documents and work with the data inside.
Zeno serves as a mediator between your knowledge base and you by providing insights from your data with references.
You can centralize a multitude of data sources into your knowledge base. Access diverse information right on ZenoChat to boost its efficiency and adaptiveness.
Here are some popular use cases among other users:
- Connecting existing Google Drive files with 1 click and working with them
- Summarizing research papers or meeting notes
- Extracting key highlights in long PDF / DOCX files
- Analyzing CSV sheets with rows of data
- Reading publicly accessible web URLs and working with all the information there
How do knowledge bases work?
Your inquiry triggers Zeno to inspect and define the aim of your instruction. Based on the objective and the relevance as primary criteria, Zeno carefully looks through the vast accumulation of knowledge, giving you substantial and relevant pieces of information that match your intended goal.
Essentially, the AI acts like a search engine and only looks for the sections that are related to your query. If it's a question about Section A, it will try to go to Section A and fetch you the relevant information so that you can access it quickly and more efficiently.
Additionally, AI models have something called a "context window" which sets a limit for how many characters they can process in one run.
Therefore, AI cannot be expected to know everything uploaded in your knowledge base. Its actions are driven by on-demand instructions, needs and requirements within the context window. That's why being specific in your message / instruction greatly enhances the output you get.
What about summarizing a knowledge base?
Asking AI to summarize a whole knowledge base is most of the time not viable and often results in dissatisfaction since it lacks the general direction of an objective to achieve.
This is very similar to expecting a good outcome from the following query:
Using Focus Search for a better precision when working with one file at a time
Using Focus Search can lead to better results if you only intend to work with one document/file. This gives the AI a general direction on where to look and what information to retrieve for you. In general, it's more effective to work with one document at a time.
You can simply select a single file in the Files dropdown menu and query information only from that specific document, rather than searching the entire knowledge base.
Example: "Give me the action items that need to be done out of this meeting."
This will give Zeno a focused scope to understand which file to work with. However, make sure that the amount of words inside the document is within the context window of the chosen model.
Keywords and specific terms
It is also possible to look for certain keywords, terms or documentation and learn more about your internal processes: you can simply enable your shared knowledge base and insert your inquiry about the information you are searching for. If there is a relevant source inside the knowledge base that can correspond to your message, Zeno will fetch it for you.
Can different sources/different systems be consulted and searched?
Yes, any source that has been uploaded to a knowledge base can be consulted and information can be retrieved.
Can connections/comparisons be made between different documents?
Yes, you can refer to multiple documents and ask Zeno to make connections between them. But keep in mind that the AI does not have an infinite memory to process a lot of information in one go.
If you are referring to way too many files with a lot of text data in one message, some of the documents might not get referred or taken into account when generating responses due to the overwhelming number of characters and insufficient context window of the models.
Therefore, it is best to split major tasks into smaller sub-tasks and work with a handful of data at a time.
How can I structure my knowledge the best?
When deciding how many files, what mix of file types, and how to process them for your knowledge bases (KBs), the key trade-off lies between having a single, comprehensive repository versus multiple, focused ones. Users often struggle with this structure.
Our Recommendation: Group by Purpose, Not Size
-
We strongly advise you to group your knowledge bases by purpose or department (e.g., "HR Policies," "Marketing Assets," "Project Alpha Documentation") rather than creating one gigantic knowledge base containing everything.
-
Why? The Trade-offs Favor Focused KBs:
- Easier Sharing & Access Control: Smaller, purpose-built KBs make it simpler to share relevant information with specific teams or individuals without granting access to unrelated sensitive data.
- Improved AI Performance: Grouping gives the AI a clearer scope of files to work with for a given query. This leads to higher accuracy and precision because the AI isn't wading through vast amounts of irrelevant information.
- Efficient Context Window Use: With focused KBs, the relevant files retrieved for a query are less likely to collectively overwhelm the AI's context window. This ensures that important files don't inadvertently get excluded or truncated.
The Context Window Challenge with Large KBs or Files
- AI models have a limited "context window" – the amount of information they can consider at once when answering a query.
- In one search query, the system might identify multiple files as potentially relevant and attempt to put snippets from them into the context window.
- The Trade-off: If your KB is massive, or if even one retrieved file has very large context (like a dense technical manual or a long transcript), it alone can consume the majority of the AI's available context. This leaves little room for information from other potentially relevant files, leading to incomplete or skewed answers.
Practical Strategy: Leverage Both Broad and Focused Searches
Understanding this context limitation leads to a more effective search strategy:
- Broad Search (Discovery): When you need to find out which documents might be relevant across a wider topic (e.g., "Find all documents mentioning Q3 product launch"), searching across a larger collection or multiple KBs is appropriate. The goal here is primarily document identification.
- Focused Search (Detail): Once you've identified key documents, or if you have a very specific question tied to a known area, it's better to perform search operations on a limited number of files or within a specific, smaller KB (e.g., "Summarize the key risks for the Q3 launch based on the Project Plan and Risk Register"). This allows the AI to dedicate its context window to the most relevant information, enabling more detailed and accurate answers.
Create your first knowledge base
1. Navigate to the Knowledge section on our app. From there, click on the Knowledge Bases tab and hit the +New button.
2. Select a name for your Knowledge Base and then upload your desired files. We support various file formats such as PDF, DOCX, CSV, and PPTX to ensure seamless integration with your existing documents.
3. Once your files have been uploaded, you can head over to ZenoChat, click on the "Knowledge" button and choose the Knowledge Base you want to work with.
Note: Keep in mind that all files are processed by TextCortex without the use of third parties. Refer to our article "How We Handle Data at TextCortex" for more information.
4. Effortlessly search your knowledge base for specific answers and seamlessly work alongside your documents in a unified interface.
Integrations (Google Drive, Microsoft OneDrive...)
You also have the option to connect third-party accounts such as Google Drive, Microsoft OneDrive and many more to come. These integrations enable you to import all of your existing data into your knowledge base seamlessly with one click.
You're ready to go!
That's it! You're now ready to harness the full power of your knowledge bases. Go ahead and ask questions to your files.
IMPORTANT: Please keep in mind that Zeno does not do a full scan of all the documents across your knowledge base (yet). At the current rate, it only retrieves the relevant sections that match the intent of your message.