Systems | Development | Analytics | API | Testing

At Koyeb, we provide high-performance serverless infrastructure for intensive applications across CPUs, GPUs, and accelerators. We take code, build it, and run it in fully isolated, secure microVMs on bare-metal servers around the world. We scale automatically when needed, down to zero when idle, with cold starts as low as 250ms. Over the past few months, we’ve been working with an increasing number of teams using Koyeb to orchestrate and run AI-generated code at massive scale.

In 2025, as AI models increasingly generate, refactor, and deploy code on their own, developers face a new challenge: how to safely run code they didn’t write. Sandboxes have become the backbone of this new workflow because they are lightweight, secure environments that let teams test, validate, and monitor code without risking production systems.

In September 2025, we brought AI Engineer to Paris for a two-day conference with 700+ attendees, 25 sponsors, over 47 talks cross 5 tracks. This is the third post in a 4-part series on the event: If you missed AIE Paris, you’ll be able to watch the replays on the Koyeb YouTube channel. Main Stage Day 1 Main Stage Day 2 If you’re not familiar, the team behind the AI Engineer has been running AI Engineer conferences in the US for the past three years.

In our previous blog post, we introduced two ways that Confluent Cloud can integrate with AWS Lambda. One option is using Lambda’s Event Source Mapping (ESM) for Apache Kafka, wherein Lambda creates a consumer group, consumes records off the provided topic, and triggers the Lambda function. The record is polled by the ESM, and the consumed record subsequently acts as the event data provided to (and processed by) the Lambda function.

On September 23 and 24, we hosted AI Engineer Paris 2025 at Station F — a one-day gathering of builders, researchers, and practitioners exploring the future of applied AI. With five talk tracks, 48 sessions, and 25 sponsors, the event brought together the best of the AI engineering community in Europe and worldwide. If you want a full recap of the key themes and takeaways from the talks, check out our event recap blog post.

At Koyeb, we run a serverless platform for deploying production-grade applications on high-performance infrastructure—GPUs, CPUs, and accelerators. You push code or containers; we handle everything from build to global deployment, running workloads in secure, lightweight virtual machines on bare-metal servers around the world.

At Koyeb, we run a serverless platform for deploying production-grade applications on high-performance infrastructure—GPUs, CPUs, and accelerators. You push code or containers; we handle everything from build to global deployment, running workloads in secure, lightweight virtual machines on bare-metal servers around the world.

Today, we're thrilled to announce the public preview of Light Sleep. Waking up from Scale-to-Zero is now imperceptible for CPU workloads with sub-200ms cold starts. A few months ago, we announced the first iteration of Scale-to-Zero on the platform to reduce idling costs. With Scale-to-Zero and Autoscaling, apps sleep and wake up automatically on demand based on requests, and scale out horizontally according to your criteria.

Today, we’re announcing the public preview of TCP Proxy — a new way to expose TCP ports publicly. Until now, services on Koyeb could only be publicly exposed via HTTP, HTTP/2, WebSocket, and gRPC protocols. TCP-based workloads were limited to private access within the mesh network for service-to-service communication. With TCP Proxy, that changes. You can now make any TCP service publicly accessible with minimal configuration.

At Koyeb, we run a serverless platform for deploying production-grade applications on high-performance infrastructure—GPUs, CPUs, and accelerators. You push code or containers; we handle everything from build to global deployment, running workloads in secure, lightweight virtual machines on bare-metal servers around the world. One of the core metrics we track is time to HTTP 200: the time between a successful deployment (excluding build) and the moment the app is ready to accept traffic.