Mistral AI, renowned for its frontier large language models (LLMs), needed to integrate with customers' file storage applications to power their cutting-edge enterprise AI search product for businesses.

But their team didn't want their engineers focused on integration-related projects, as that would distract them from working on their LLMs.

The head of engineering at Mistral shares more context on this predicament:

“Building integrations in-house would take our engineers hundreds of hours initially and several more each week for maintenance."

As they evaluated different options for implementing and maintaining file storage integrations, they landed on Merge for a few reasons:

• Using Merge’s Direct File Download endpoint to directly access files without storing them on Merge’s servers

• Complying with critical security frameworks, such as GDPR, HIPAA, ISO 27001, and SOC 2 Type II

• Encrypting data both in transit and at rest

• Offering a single-tenant environment to allow users to host data in the AWS region of their choice

• Enterprise-level customer success and support

Here’s how the file storage integrations work:

1. An admin for the file storage solution would log into the enterprise AI search product and authenticate a file storage connection via Merge Link—a UI component that’s embedded into their app.

2. Once the customer admin completes the guided authentication steps successfully, Merge traverses the file storage structure and normalizes it according to Merge’s Common Models. This allows the LLM to embed the data accurately before it gets added to a vector database that’s operated by Mistral AI.

Throughout this process, Merge doesn’t download and store file contents. Merge’s File Storage Common Model serves as a metadata representation of the file object, storing information about the file type, size, location, and which users can access it. In addition, Mistral AI uses Merge’s ACLs to ensure users only receive outputs from the files they have permission to access.

The collected data is refreshed via Merge as frequently as every 5 minutes to keep the file contents in the vector database accurate and complete.

3. The customer’s employees can input questions in the enterprise AI search product, which prompts the AI chatbot to retrieve the relevant context in the vector database and use it to generate an output.

Merge allowed Mistral AI to launch their integrations several months faster than they could by building them in-house.

Mistral AI’s head of engineering adds more context:

"We had an aggressive timeline for launching our enterprise AI search product. Merge moved quickly in supporting the file storage integrations we needed, which allowed us to meet our target launch date."

Merge has also allowed Mistral to support integrations that use access control lists (ACLs). Their head of engineering explains why this is critical: “Merge’s ability to incorporate access control lists (ACLs) across their file storage integrations ensures that sensitive data stays secure and that our customers keep compliant with key data protection regulations, like GDPR.”

Finally, Merge has allowed their engineers to save countless time.

According to their head of engineering:

"With Merge, we spent just a few hours integrating with their Unified API, and we can now easily add file storage integrations without worrying about ongoing maintenance—saving our engineers significant time.”

We look forward to helping Mistral AI power a leading enterprise AI search experience.