Ai

Daytona 0.174.0 Adds NVIDIA GPU Support for Sandboxes, New Terminal Actions

Daytona 0.174.0 Adds NVIDIA GPU Support for Sandboxes, New Terminal Actions

Daytona just shipped version 0.174.0, and it's a big one for anyone running GPU-intensive workloads. The headline feature? Native NVIDIA GPU support for sandboxes. This isn't just a minor tweak — it's a fundamental shift in how Daytona handles AI and machine learning development environments.

Let's dig into what actually changed, because there's more here than just GPU acceleration.

What Changed

The core addition is the NVIDIA GPU-enabled runner — a new runner type that lets you attach physical GPUs to sandboxes. This is critical for AI developers who need real hardware for training or inference, not just CPU-bound simulations. The PR (#4671) touches both the API and runner layers, meaning it's not a bolt-on feature but a first-class capability.

Second, the dashboard now includes a sandboxes terminal action (#4641). You can directly open a terminal into a running sandbox from the web UI. It's a quality-of-life improvement that cuts out the middleman — no more SSHing manually or digging for connection strings.

On the docs side, there's a new structure for llms.txt files (#4634) — something that helps LLMs better index and understand project documentation. And, unsurprisingly, the docs now include a guide for setting up sandboxes with NVIDIA GPU devices (#4672).

A single fix rounds out the release: the CLI now handles snapshot creation more reliably (#fixed). Details are sparse, but it's likely a stability fix for a common pain point.

Why It Matters

This release positions Daytona as a serious contender for AI development workflows. Without GPU support, sandbox platforms are just fancy containers. With it, they become actual workstations for data scientists and ML engineers. The timing is interesting — as more companies move toward ephemeral, cloud-native development environments, having hardware access baked in is a differentiator.

I'd argue the terminal action is almost as important. It's one of those features you don't appreciate until you need it. If you're debugging a sandbox that's acting up, clicking a button to get a shell is way faster than copying a command. It's the kind of polish that separates good tools from great ones.

The llms.txt docs update is forward-thinking. As AI coding assistants become more common, structured documentation will help them give better answers. It's a small investment that could pay off big as LLM usage grows.

None of this is revolutionary on its own, but taken together, version 0.174.0 shows Daytona is listening to its users — especially the AI crowd. If you've been waiting for GPU support to move your workflows to Daytona, this is your cue.

Official Source: https://github.com/daytonaio/daytona/releases/tag/v0.174.0