1. Overview#
The Search Block allows you to fetch information from documents to make it available to AI. It creates a temporary data source available during publication creation.The principle of this block is to fetch document excerpts based on a question.For example, to the question "What services are offered?" the Search Block will fetch excerpts from the document that mention the offered services and return them as is.š Reference Integration: You will use the Search Block by nesting it in AI blocks so that AI can rely on these excerpts. This provides dynamic, relevant, and adapted reference elements.
Complete workflow understanding#
Before manipulating the Search Block, it's important to understand its overall functioning.1
š File upload setup
First, a "File Upload" step in the Wizard will allow the App User to add files. These files will be associated with a File Attribute.
2
š¤ File transmission
These files will be transmitted to the Search Block through the File Attribute when the Search Block is executed.
3
āļø Processing time
The Search Block will take from a few seconds to several minutes to divide these files into chunks and store them.
4
š Search availability
From this point, files are available for search by this Search Block and all those using the same File Attribute during the publication journey.
5
š¤ AI integration
By executing AI Blocks that are parents of Search Blocks during the Wizard, the Search Blocks will return excerpts for the AI to rely on and generate its content.
6
š Custom publication
The publication is created on demand by relying on reference excerpts from documents deposited by the App User.
2. Specifications#
Block Parameters#
In the Search Block parameters, you will need to specify the File Attribute that will be used.š File Attribute Setup
This File Attribute must be associated with a File Upload Step in the Wizard so that the App User can deposit the file(s) that will feed this File Attribute.š Results Configuration
You can then choose the number of results returned by the Search Block search, up to a maximum of 100.š” Optimization tip: The search will be based on the question asked, so it's preferable to work on the question formulation rather than setting a large number of results to avoid returning a large quantity of content that could confuse the AI with information overload.The shorter the document, the fewer results you'll need.
Question configuration#
Once the block configuration is done, you can nest a Paragraph Block to write the Question that will be asked to the Search Block.On the same principle as a Paragraph Block child of an AI Block becomes its Prompt, the Paragraph Block child of a Search Block becomes its Question.AI Block integration#
You can then nest the Search Block in an AI Block. The content it returns during execution will serve as a reference for the AI block's Prompt.3. Summary and best practices#
Complete setup checklist#
To recap all the steps necessary for proper Search Block functioning:š§ Setup requirements#
Create a File Attribute and use it in multiple distinct Search Blocks
Adapt the number of results according to the question and document length
Each Search Block must have a simple and explicit question
š¤ Integration requirements#
Each Search Block must be nested in an AI Block for document excerpts to be exploited
Structure properly your AI-Search Block hierarchy
Test thoroughly with different document types
3.1 FAQ#
ā AI repeats its prompt or generates content unrelated to my documents, why?
This can occur if the AI simply has no reference content to use. In other words, if the Search Block returned no results.You can also verify that the Search Block returns results by temporarily placing it as a parent block to see its content in the publications interface. If it returns no results, it means the document you deposited cannot be processed (because it's too complex or too large).š Search Block returns different results / AI never talks about the same things when relying on my Search Block as reference, why?
The question asked in the Search Block probably needs reformulation. Make sure it's simple, short, and explicit. The Search Block isn't intelligent per se - think of the question as you would pose it in a Google search by choosing your keywords well. It's the keywords, question clarity, and precision of terms to find that will determine question relevance and result quality.You can also adjust the number of results: reduce it if the search targets a specific document element, increase it if the document is very long and may contain relevant information on the subject at many points.š¤ Is it a good idea to nest multiple Search Blocks in a single AI Block?
Yes, as long as you frame the amount of information that will be delivered to the AI. Keep in mind that all excerpts will be provided to your Prompt, which can create an endless Prompt and confuse the AI.Best Practice: Put multiple Search Blocks in an AI Block that will synthesize the recovered excerpt elements, then use this AI Block (the synthesis) to feed other AI Blocks. You'll have relevant and calibrated content to use as reference in your other AI Blocks, without having to use numerous Search Blocks and their results each time.šÆ Key insight: Keep in mind that the Search Block returns document excerpts, not an entire document. The AI will rely on these excerpts as reference. If you return too many that don't correspond, it will be lost; conversely, if you don't return enough, it will lack precision.
šÆ Question quality
Craft precise, keyword-rich questions like Google search queriesš Result balance
Balance excerpt quantity with relevance - more isn't always betterš Iterative testing
Test different question formulations and result counts to optimize performance