Meilisearch: a superfast and intuitive open-source search engine in SaaS mode

Meilisearch designed its search engine with developers in mind right from the start, distributing it as an open-source product. This developer-friendly approach is a powerful mechanism to encourage adoption: “ GitHub is our main entry point,” emphasizes Thomas Payet, co-founder and COO. Unlike more complex search engines that are difficult to deploy or closed by default, Meilisearch is committed to transparency and technical accessibility. Each release is tested in real-world conditions by thousands of developers, who provide immediate feedback. This community dynamic accelerates product iteration while enhancing the solution's reliability. “It's a form of collective agility that enhances the strength of the Meilisearch ecosystem,” explains Thomas Payet.

With the launch of its SaaS offering in 2022, Meilisearch faced two challenges: ensuring flawless uptime while building an infrastructure in line with the security standards expected by large accounts. However, its infrastructure hosted by DigitalOcean had several limitations: restrictions on storage volume per instance, late maintenance notifications, and difficulty scaling dynamically. “With our previous provider, we had to regularly contact tech support to get more machines or unlock a feature. With AWS, there are no limits; we can scale infinitely,” summarizes Thomas Payet.

Migrating to AWS was consistent with the previous architecture, as Meilisearch was already using Kubernetes in managed mode at DigitalOcean. On AWS, the company is deploying Amazon EKS (Elastic Kubernetes Service), which offers a similar solution but is more reliable and better integrated.

Meilisearch relies on Amazon EC2 instances configured to track demand growth for computing and storage. Thanks to this infrastructure, the platform can absorb customer growth smoothly while maintaining high performance levels.

Choosing AWS is also in line with the company's desire to structure itself around high security standards (SOC 2, ISO 27001). With the help of AWS partner Skale-5, Meilisearch is defining an architecture designed for resilience and compliance: access is secured via a single entry point, each environment is isolated to limit risks, and data is stored on automatically replicated volumes (EBS) to prevent any loss should an incident occur.

With this foundation, Meilisearch can offer solid SLAs to its customers and meet their expectations. “Our infrastructure being hosted on AWS, even our most demanding customers know they can trust us,” summarizes Thomas Payet.

For Thomas Payet, the high quality of support provided by AWS was a decisive factor: “I was pleasantly surprised by the human aspect of the support provided by AWS. Every step of the way, we were put in touch with the appropriate people, who were proactive and genuinely sought to understand our needs.” A relationship of trust was built up through numerous exchanges with AWS, as Thomas Payet points out: “Our account manager isn't trying to sell us more; he's there to help us succeed. He's already saved us 10 to 20% on our invoice.”

In a world where the use of generative AI models is skyrocketing, Meilisearch intends to become a key building block in Retrieval-Augmented Generation (RAG) architectures. The aim is to help companies leverage their internal data (product catalogues, knowledge bases, FAQs, etc.), whether to simplify document search or deploy AI agents. “Nowadays, everyone uses ChatGPT. But how can we make these tools equally relevant to a company's data? That's where Meilisearch comes in,” explains Thomas Payet. With annual revenue now exceeding €1 million, Meilisearch intends to expand its customer base and structure its commercial approach. To this end, Meilisearch can count on AWS to guarantee the reliability, scalability and security essential to its goals.

Founded in 2018, Meilisearch is developing an open-source search engine that is easy to integrate and superfast (<50 ms). Used by companies such as Hugging Face, GoTo and Bookshop.org, it is actively supported by a community of over 50,000 developers.