How Qovery uses Qovery to speed up its AI project
At Qovery we use our own products every day. We believe it is the fastest way to build empathy for our users and push innovation further. Whether it’s a frontend engineer using Preview Environments to test a pull request in isolation or our backend team refining our API, Qovery is the heartbeat of our development lifecycle.
But today is another story, we’re taking you behind the scenes to show how we use Qovery to speed up our new AI initiative.
The Challenge: Speed vs. Complexity
We’ve been busy developing our AI DevOps Copilot, an AI assistant designed to redefine how developers interact with infrastructure. Our goal was simple: help developers and DevOps teams resolve daily operational headaches so they can stay focused on strategic, high-value work.
To build this, we needed the right talent. We found an incredible AI expert to lead the initiative. He is a master of Large Language Models (LLMs) and data science, but like many specialized developers, his expertise isn't in infrastructure, Kubernetes clusters, DNS configuration, or deployment management.
In a typical setup, this engineer would have spent weeks waiting for a Senior Reliability Engineer (SRE) to:
- Set up CI/CD pipelines.
- Configure Kubernetes and Helm charts.
- Manage DNS and SSL certificates.
- Provision databases and monitoring.
Weeks of back-and-forth, copy/pasting of YAML configuration to put in place even before his code can reach its target, clients.
How He Plugged the AI Stack into Qovery
For our AI lead, the challenge wasn't just writing the code; it was ensuring that code could talk to the right data and scale on demand. Here is exactly how he "plugged" the AI stack into Qovery:
- Connecting the Brain (LLM & QDrant):
The core part of the application containing the agent and MCP server is built using our automatic docker image builder pipeline and request on-demand GPU nodes during deployment. To function, it needs to connect to databases; while Qovery provides ready-to-deploy SQL databases, we initially lacked a solution for vector databases like Qdrant to store model embeddings. To solve this, we leveraged Qovery’s Helm deployment system to deploy the official chart of QDrant, without the need for our AI lead to touch any CI/CD pipeline.
- Automated scaling with KEDA
Since some operations of the agent can require a lot of memory, we enabled our beta autoscaling based on KEDA. This ensures we have the correct number of replicas to handle the load and set a readiness probe to stop serving traffic once a threshold is reached.
- Managing the "State" with Terraform
As the product grows, so do its dependencies; our AI lead soon needed to store data into object storage. Instead of opening a ticket to our SRE to handle the creation of the S3 bucket, he used Qovery’s new Terraform integration to manage the lifecycle of S3 bucket directly within him environment, making him completely autonomous.
Empowering the Individual
With Qovery, our AI lead pushed his application to production in a single day. He didn't just "deploy code"; he deployed a production-grade system with all the "safeties" an SRE requires, handled automatically by the platform:
- Automatic Rollbacks: Instant recovery if a new model version fails liveness probes.
- Deterministic Builds: Using Docker to ensure the AI environment is identical from local to prod.
- Full Observability: Real-time monitoring to observe the live state of the agent's performance.
Conclusion: Infrastructure as Leverage
We are really proud of our new AI DevOps Copilot initiative, It proves what we’ve always believed: Infrastructure should be leverage and AI is a big advantage for the future
Qovery has become the platform we envisioned: a place where developers are empowered to move at the speed of their ideas, and DevOps teams know the guardrails are baked into the system.
Check out our AI Copilot or try Qovery yourself!
.svg)
.svg)
.svg)
Suggested articles
The cluster coming up is the easy part. What catches teams off guard is what happens six months later: certificates expire without a single alert, node pools run at 40% over-provisioned because nobody revisited the initial resource requests, and a manual kubectl patch applied during a 2am incident is now permanent state. Agentic control planes enforce declared state continuously. Monitoring tools just report the problem.
The instinct when setting up Kubernetes observability is to instrument everything and send it all to your APM vendor. That works fine at ten nodes. At a hundred, the bill becomes a board-level conversation. The less obvious problem is the fix most teams reach for: aggressive sampling. That is how intermittent failures affecting 1% of requests disappear from your monitoring entirely.
Scaling your deployments to zero is only half the battle. If your cluster autoscaler does not aggressively bin-pack and terminate the underlying worker nodes, you are still paying for idle metal. True environment sleeping requires tight integration between your ingress layer and your node provisioner to actually realize FinOps savings.
The structure, table, tool list, and code blocks are all worth keeping. The main work is fixing AI-isms in the prose, updating the case study to real metrics, correcting the FAQ format, and replacing the CTAs with the proper HTML blocks. The tool descriptions need the "Core strengths / Potential weaknesses" headers made less template-y, and the intro needs a sharper human voice.
For years, Red Hat OpenShift has been the safe choice for heavily regulated, on-premise environments. It operates as a secure fortress. But in the public cloud, that fortress acts as an expensive prison. Paying proprietary per-core licensing fees on top of your standard AWS or GCP compute bill is a redundant "middleware tax." Escaping OpenShift requires decoupling your infrastructure from your developer experience by running standard, vanilla Kubernetes paired with an agentic control plane.
Agentic Kubernetes resource reclamation is the practice of using an autonomous control plane to continuously identify, suspend, and delete idle infrastructure across a multi-cloud Kubernetes fleet. It replaces manual cleanup and reactive autoscaling with intent-based policies that act on business state, eliminating the configuration drift and cloud waste typical of unmanaged fleets.
Kubernetes focuses on container orchestration, but the reality on the ground is far less forgiving. Provisioning a single cluster is a trivial Day-1 exercise. The true operational nightmare begins on Day 2. Teams that treat multi-cloud fleets like isolated pets inevitably face crushing YAML configuration drift, runaway AWS bills, and severe scaling bottlenecks.
.webp)