Transcript
We have a great show for you today. We will be doing a lot of building with agents that live inside your database. It is a cool way to build agents and we are excited to explore this.
There are big implications for those of us who are building Softr apps. We will have agents in the database that can do a lot of heavy lifting for us. It goes beyond just building apps.
Now, managing your data and having an agent within your database can help with finding, fetching, summarizing, extracting, and translating data. It keeps your data hygiene in a better spot to power your automations and interfaces.
[.blog-callout]
Note from Softr: A great way to get started is by using the AI co-builder. It builds complete apps, pages, or database schemas instantly, while still letting you tweak things manually later.
[.blog-callout]
Softr is one of the simplest ways to build applications where your data is safe. It is moving towards becoming an AI platform, and the most exciting feature released this year is AI agents in Softr Databases.
We are going to build four different things today. We will work with customer support use cases, data extraction from files, translation, and lead enrichment. We want to use these agents in many different ways.
Let's imagine you have a customer support queue where tickets are saved into the database. When a ticket comes in, we want to detect the language and translate it to English to standardize it.
From there, we want to detect the sentiment of the customer support ticket to see how happy people are. We will also have AI help us find a ticket priority to organize our queue properly.
[.blog-callout]
Note from Softr: While Softr integrates with 17+ external data sources, Softr Databases is the powerful, native way to manage data directly within Softr for maximum performance.
[.blog-callout]
Softr Databases allow you to work with interfaces or use the database as a standalone product. It is very similar to Airtable, but we built it to be a scalable database for building at scale.
If you want to pull this database into n8n, Make, or Zapier, we have integrations for those platforms. Our API limits are very high, making it great for programmatic building and easy to use.
To build our first agent, we select the AI agent field and call it language. We can choose which model we want to use, including the latest models like Sonnet 3.5 or GPT-4o.
We enter a prompt using a variable to reference the customer message field. We ask the AI to detect the language and report it back. We can also use AI to improve the prompt for us.
Web search allows the LLM to search the live web in addition to its own data. This makes the data you are finding very relevant, which is great for use cases like lead extraction.
We have a lot of control over when these agents run. You can trigger them manually, when a new record is created, or when a record is updated. This ensures they only run when you want.
We can also add an English translation field. If the customer message is not in English, we can have a conditional filter so the agent only runs when the language detected is not English.
[.blog-callout]
Note from Softr: To automate your business operations even further, check out Softr Workflows. It allows you to keep your logic close to your design without relying solely on third-party automation platforms.
[.blog-callout]
Next, we can add a customer sentiment score. On a scale of 1 to 100, we can ask the AI what the sentiment of the message is. This helps us track how customer happiness is trending.
We can even have the AI suggest a response. We prompt it to be polite and professional, using Softr documentation as knowledge. By turning on web search, it can find relevant technical docs.
For document extraction, we can upload invoices and extract the ID, date, amount, and merchant name. We just create a field for each data point and ask the agent to extract it.
We can set conditions so the extraction only runs when the invoice file field is not empty. This saves credits by ensuring the agent doesn't run on empty rows.
We also have the Ask AI feature, which adds a chatbot that your users can use to ask questions about their data. This is very helpful for client portals or internal tools.
[.blog-callout]
Note from Softr: Building an ERP template or a client portal is much faster when you use native features like Ask AI to let users query their own data.
[.blog-callout]
With web search, we can also do competitor research. We can ask an agent to find the average market price for a product by searching the web and returning the value in a currency field.
If a native block doesn't fit your exact needs for displaying this researched data, you can use the Vibe-Coding block. You simply prompt for the component you want and it connects to your database.
Softr has grown from an interface product to a full-stack platform. We have Softr Databases, Softr Workflows, and AI-infused features across the whole suite to help you build professional software.
You can use these agents for extraction, summarization, translation, and enrichment. The form factor in the database makes it easy to keep your data hygiene at a high level.
