Why Does My AI Agent Return Only Partial Location Data from My Supabase Vector Database?

Below are some suggestions to troubleshoot and improve your setup:

1. Check Your Result Limit

Most implementations—including the Supabase Vector Store node in n8n—come with a default limit on the number of results returned. In your case, your agent may be set up (or the underlying query may be configured) to return only 4–8 matches. You can adjust this limit with the parameter (often labeled Limit) on the node or in your SQL query. Increasing this value should help retrieve all matching entries.

2. Consider Data Structure

Storing your JSON objects as a single string can work, but it may not be the most optimized approach for semantic search. If you use structured fields instead, you can improve both the quality of the search results and the ease of filtering by location. For example, storing key property information like "Localidad" or "Provincia" in separate columns can make it easier to perform targeted queries.

3. Token/Context Window Considerations

When using an AI Agent, keep in mind that the agent’s response may be constrained by token limits. Even if you retrieve more entries, the response might omit some details if the total token count exceeds what the model can return. If your ultimate goal is to process or extract all the data, consider:

• Returning a list of identifiers or references.
• Calling a separate action (or pagination) to fetch the complete data when needed.

4. Optimizing the Semantic Search Setup

For better search results with AI and vector search:

• Ensure your embeddings capture the location and other spatial information well.
• Validate that the summarization you’re using includes enough detail for location queries.
• Experiment with different embedding models or generation configurations (as provided in the Supabase toolkit) to see if they better capture the properties’ nuances.

Example: Adjusting the Limit in n8n’s Supabase Node

If you are using the Supabase Vector Store node, look for a parameter called Limit under the “Get Many” settings:

YAMLCopied
### Get Many parameters

* **Table Name**: Enter the Supabase table.
* **Prompt**: Your search query, e.g. "lands within [location]".
* **Limit**: Set this to a higher number (for example, 50) to retrieve more matches.

Updating this setting should help your agent return all entries related to the location term.

Summary

• Increase the retrieval Limit parameter in your Supabase/n8n node.
• Consider structuring your data into discrete columns for more effective targeting.
• Be aware of token constraints in AI Agents and plan for pagination or separate follow-up queries if needed.
• Experiment with embedding generation to ensure location details are well reflected in your vector representations.

These adjustments should help you retrieve a more complete dataset using your AI Agent.