The most powerful S3 API ever? Introducing the Prompt API.

The object storage world to date has been defined by the S3 API concepts of PUT and GET. The world in which we live now, however, requires more. Given that MinIO has more S3 deployments than even Amazon, it fell to us to come up with the next great S3 API. 

That new API is the Prompt API and it very well may become the most powerful extension of the S3 API ever created. It will change the PUT and GET paradigm to the PUT and PROMPT paradigm and with it, how users and applications interact with the data. It is available in the new AIStor.

At the most basic level, the MinIO promptObject API lets users or applications talk to unstructured objects as if they were talking to an LLM. That means you can ask an object to describe itself, to find similarities with other objects and to find differences with other objects. It will lead to an explosion of applications that can speak directly to the data residing on MinIO. More importantly, the promptObject API is effectively transparent to the user/application. No prior knowledge of RAG models, vector databases or other AI concepts is required. The promptObject API works out of the box with multi-agentic architectures where orchestration is built in to work with small-scale, domain AI specific models.

This follows the core mantra of MinIO around simplicity. It should not be complicated or expensive for an organization to get basic AI working - you just need to be able to point the promptObject API at your data. 

The promptObject API delivers its capabilities as an API. It is not a standalone application or an extension to our Console (although a user interface for it is available in the console) but rather an extension to our SDKs (we support all the major programming languages). In short, promptObject can enrich existing applications or build new ones. 

Let’s start with a demonstration as to how the promptObject API works before turning to the details underpinning it.

The Makings of an Expensify-Like Application

The promptObject API works with any unstructured object. Images are one such unstructured object and expense receipts are a perfect example. Here is one:

There are two ways to “talk” to this object. One is directly through the Console using the promptObject box and the other is through the Jupyter Notebook of a developer. We are going to start with the Jupyter Notebook because it shows the image we are working with and will provide a little more context to the reader.

Here is an example of the Notebook. We are browsing objects from AIStor inside Notebooks and talk to the object using prompt_object 

The same functionality is available in the Console. The object browser is used to manually verify the image if needed.  

Here we ask the object how many people came to dinner.

In either case, the user can ask almost any question about the receipt. What was the average check size, what city is it in, what is the image at the top, what was the most expensive dish? 

MinIO runs a multi-modal LLM on the backend and takes care of everything. It is totally transparent to the IT user or application developer (but naturally open to the data science team for inspection). This does require GPUs, but the team could get started with just one.

Querying multiple objects at once is also supported. In this case, the user identifies three receipts and ask the promptObject API to find two differences and two similarities. 

The possibilities are effectively endless both from a user perspective as well as from an application perspective. One easy but incredibly powerful example is the concept of appending tags to objects. Often, objects come into the AIStor with little descriptive labels. They are generated from machines or from audio or video. Using the promptObject API and a little code, an application can ask each new object to describe itself, to summarize itself, to give specific data about itself. It can then append that information to the object itself - thereby making search better (see the Catalog feature), or other applications more effective. 

In the GIF below you will see the user adding new tags for multiple objects. Those tags automatically detect if the content has PII, would be subject to GDPR or has a sensitivity level that would restrict it to internal use only. This can be combined with Catalog to search for objects for specific tag(s) and their corresponding values.

Keep in mind, the rules on metadata still apply here. A third party metadata database doesn’t scale. You need to adopt architectures where the metadata is atomic to the object (e.g. AIStor). Nonetheless, more metadata/tags is a good thing and having an AI label data as it arrives is a total game changer.

A Game Changer for the Business

The business benefits of this should be readily apparent. Data is the core value in the organization. Unstructured data represents north of 90% of all data in the enterprise. Knowing more about that data and being able to easily interact with that unstructured data - without having to have RAG or vector database skills is massive. Application developers can develop whatever the business needs. The data is stored as it has always been - in MinIO. The business team comes up with a concept, they can test it themselves in the Console or describe it to the application team. The application team can build something - likely within hours. The goal of AI is to extract more business value from the data. Here are some of the benefits to the enterprise powered by promptObject:

1. Transforms Data into an Immediate Resource for Business Insights. No specialized skills required. 

Traditionally, enterprises store vast amounts of unstructured data, but mining actionable insights from it has required specialized data teams, costly analytics tools, and complex ETL (Extract, Transform, Load) processes. The promptObject API could eliminate many of these bottlenecks by making it possible for anyone within the enterprise to query the data directly and naturally.

This effectively turns data storage into an active resource rather than a passive archive. Enterprises can directly ask questions of their data and get valuable responses, whether it's about customer behavior, compliance, or operational insights, without needing separate data pipelines or expert intervention. This capability speeds up decision-making and enables proactive data-driven actions across departments.

2. Lowers the AI Skill Barrier, Making AI Accessible Across the Organization

By abstracting away complex AI concepts (e.g., retrieval-augmented generation, vector databases), promptObject makes it feasible for non-technical teams to leverage AI-driven insights. For example, compliance officers could verify document compliance, customer service can analyze interaction histories, and marketing teams could assess campaign performance directly from storage. This democratization of AI drastically reduces the dependency on specialized data science teams, making AI tools more available and accessible to business users. Enterprises can enable self-service data analysis across departments, which not only enhances productivity but also allows teams to be more innovative with data. This is particularly valuable for enterprises that might otherwise struggle to hire or retain AI talent, especially in industries outside of tech.

3. Reduces Infrastructure and Operational Costs Related to Data Analytics

Most enterprises that engage in AI-driven data analysis require additional infrastructure, such as data warehouses or separate AI processing platforms. By embedding AI-driven querying directly within storage, enterprises can potentially eliminate or reduce reliance on additional data processing layers, leading to streamlined infrastructure and reduced operational overhead.

This will result in significant cost savings on both infrastructure and/or cloud fees. Instead of needing dedicated AI platforms or third-party analytics solutions, enterprises can utilize promptObject for in-situ analysis, simplifying their data architecture and consolidating costs around a single, versatile storage platform.

4. Accelerates Time-to-Insight, Enabling Real-Time Operational Agility

Traditional data processing workflows involve numerous steps, from data retrieval and cleansing to analysis. The promptObject API allows enterprises to bypass these steps by interacting directly with the data in its stored format. This is particularly valuable for real-time or near-real-time insights, where the lag associated with traditional data processing is too slow.

Faster access to insights means enterprises can act on data when it’s most relevant, which is crucial for time-sensitive applications. For instance, manufacturers can monitor equipment data for predictive maintenance, finance companies can detect fraud faster, and retailers can adjust pricing strategies in real time. This real-time agility improves competitiveness and responsiveness across sectors.

5. Enhances Compliance and Governance Capabilities

With regulatory requirements on the rise, many enterprises struggle to maintain control and visibility over their unstructured data. The promptObject API’s ability to query data directly enables compliance officers to identify sensitive information, verify document integrity, or audit data usage more effectively.The API empowers compliance and governance teams to better manage regulatory risks without needing extensive data workflows. They can perform on-demand audits or compliance checks directly within storage, greatly improving data governance practices and reducing the risk of non-compliance.

6. Fosters Innovation by Reducing Experimentation Costs

Data-driven experimentation, such as testing new products or exploring customer insights, typically incurs costs related to data extraction, model training, and infrastructure. The promptObject API makes experimentation more accessible, allowing teams to query data with minimal preparation and no need for complex models.

With lower costs and barriers to data experimentation, enterprises can foster an innovation culture by encouraging teams to explore data insights freely. This could lead to faster prototyping, better customer understanding, and innovative solutions derived from previously untapped data.

What’s Next

Today, promptObject works with any unstructured object. This includes images, PDFs, GIFs etc. Longer form video still needs some optimizations. We need to add support for audio as well. 

Looking Forward

We are not generally the types of people to overstate things. We generally lead with the product and let it do the talking. It has worked out quite well for us over time. Having said that, this feature feels special. We have expanded its capability in the the time we have written this blog post and come up with multiple new applications of the API.The more you do with it, the more you think about new things. We suspect this is what the team at OpenAI felt when they first grasped what they were dealing with. Clearly these are very different in terms of magnitude, but for storage types, they will immediately grasp the magnitude of the Prompt paradigm. We are always excited to see how people push MinIO, but we are extra excited for this feature. If you want to go deeper, we can arrange for that. The demo will blow you away.