Migrating from Neptune.ai
After Neptune.ai: an experiment tracker you run yourself.
Neptune's hosted service ended on 5 March 2026. If you exported your runs, EUVYRA can take them in. It runs on your own infrastructure, so your experiment history stays under your control.
What happened to Neptune.ai
- On 3 December 2025, Neptune announced that it had entered into a definitive agreement to be acquired by OpenAI, subject to closing conditions, and closed new sign-ups and trials.
- The hosted service shut down on 5 March 2026 at 10:00 PST. The API, and with it the export, was available until then, and deletion of hosted data began at shutdown.
- For self-hosted customers, Neptune's Helm chart and container image repositories were removed on 8 March 2026.
- Neptune published Neptune Exporter, a command-line tool that downloads runs to local Parquet files, with artifacts and file series alongside, and loads them into other platforms, MLflow and Weights & Biases among them.
These dates come from Neptune's transition hub. If you still need help with an export, Neptune's own channels are the place to ask.
Where your Neptune work goes in EUVYRA
| What you did | In EUVYRA |
|---|---|
| Log parameters, metrics and system metrics | Runs with configuration, metrics per step, GPU and CPU load, git commit and package versions |
| Compare runs | Charts across runs, run comparison and saved workspaces |
| Log images, series and files | Images, audio, video, histograms, plots, text and versioned artifacts |
| Keep track of models | Model versions with model cards, artifact digests and lineage, released through approvals |
| Track runs without network access | Offline mode and euvyra sync |
On top of tracking, EUVYRA runs the training itself as jobs on your GPUs, serves approved models through a gateway, and keeps the evidence that the AI Act and your auditors ask for. Everything about training
How to move your runs
EUVYRA has no reader for Neptune's export format. The bridge is MLflow's file store, which Neptune Exporter can write and euvyra import mlflow reads. We have not tested this chain against a real Neptune export, so compare run counts afterwards.
Start EUVYRA and sign in from the CLI
Run it with Docker Compose or Helm, create an API key under Settings, API keys, and store it locally.
pip install euvyra euvyra login --host https://euvyra.example.eu --api-key euv_...Load your export into a local MLflow file store
If you exported with Neptune Exporter, its MLflow loader writes to an MLflow tracking URI. A local directory as the tracking URI produces the
mlruns/folder the EUVYRA importer reads. Recent MLflow versions use a database by default, so make sure the target is a directory.# remaining options as described in Neptune Exporter's README uv run neptune-exporter load --loader mlflow \ --mlflow-tracking-uri ./mlruns ...Import into EUVYRA
The importer reads the folder directly, without MLflow installed.
euvyra import mlflow ./mlruns --project chest-xrayPoint your training code at EUVYRA
Replace the logging calls in your training loop. From here on, new runs land in EUVYRA directly.
import euvyra with euvyra.init(project="chest-xray", config={"lr": 1e-3}) as run: for epoch in range(epochs): run.log({"train/loss": train_one_epoch()}, step=epoch) run.log_model("model.pt")
What does not carry over
euvyra import mlflow copies run names, parameters as configuration, metrics with their steps and timestamps, tags, notes, git information, status and start and end times. It does not import files, images or model registry entries. Log the ones you still need as artifacts of a run, or keep them in your own bucket and reference them from EUVYRA.
Questions about leaving Neptune.ai
Can EUVYRA read Neptune's export files directly?
Not today. EUVYRA imports from MLflow's file store and from Weights & Biases. Neptune Exporter can load into MLflow, which makes MLflow the bridge.
Is EUVYRA a hosted service?
No. You run it yourself with Docker Compose or Helm, on your own servers or in any European cloud, with PostgreSQL and object storage that you control.
Can we still export from Neptune's hosted service?
According to Neptune's transition hub, the API was available until the shutdown on 5 March 2026, and deletion of hosted data began at that point.