InstantML

What InstantML is

InstantML is an experiment tracking product for recording machine learning runs, comparing their metrics, reviewing artifacts, and exporting run data. It is available on the web and through an API, with a Python SDK whose quickstart calls for Python 3.11 or newer. The SDK provides init, log, and finish calls, and buffers entries so it can continue spooling them offline if the network drops.

Charts cover streamed scalar series, grouped averages, smoothing, zoom, and hover readouts. Configs, metrics, tags, artifacts, histograms, and media can be queried. The product lists PyTorch, JAX, and TRL alongside its Python SDK, and supports imports from W&B, MLflow, and Neptune JSON. During a migration, it can also shadow W&B scalar logs. Artifacts and checkpoints can be stored in a customer's S3 or R2 storage. Premium offers an optional BYOC GCP ClickHouse setup for tenant product data, while artifact bytes remain in InstantML-hosted object storage. Plans range from Free to Pro at $199 a month and Premium at $699 a month; enterprise or custom terms cover options such as SSO/SAML, VPC or self-hosted deployment, custom retention, compliance, dedicated support, and unusually large data footprints.

Who it suits

InstantML suits machine learning teams that need to log, compare, and query experiment runs, including teams migrating from W&B, MLflow, or Neptune. Its Python SDK, offline logging, and customer-storage options also fit workflows that need run logging to continue through network interruptions or use S3 or R2 for artifacts.

Limits

The Free plan includes 2 seats, 2 GiB hosted storage, 1M metric points, 500k API requests per month, 2 projects, and 100 runs. Free and non-billable organizations are blocked at their monthly API request allowance. Pro and Premium meter API requests and storage overage.

Getting started

1. Open the InstantML website.
2. Create or choose a plan: Free, Pro at $199 a month, or Premium at $699 a month.
3. For SDK use, follow the Python quickstart with Python 3.11 or newer.
4. Use the SDK's init, log, and finish calls to record runs.
5. Import W&B, MLflow, or Neptune JSON if migrating run data.

Verdict

Choose InstantML if your team needs experiment tracking with queryable run data, migration imports, and offline buffering in its Python SDK. The Free plan covers smaller work, while Pro starts at $199 a month; teams needing SSO/SAML, VPC or self-hosted deployment, custom retention, compliance, or dedicated support can look at enterprise or custom terms.