Determined AI

license

Apache-2.0(OSI

source

public(determined-ai/determined is the open platform)

core features withheld

yes — master/internal/license/license.go ships an IsEE() check returning true when a license key and public key are present, so an enterprise edition is unlocked by a key rather than shipped in the open tree

Determined's source is Apache-2.0, but the repository itself contains the gate: a license.go check unlocks an enterprise edition when a valid key is supplied. That is functionality withheld by a key rather than a separate product sold alongside the open core.

• https://github.com/determined-ai/determined recorded 2026-09-15

  Repository page for determined-ai/determined: Apache-2.0 license, public and unarchived, last push 2025-03-20. Establishes the license, that the source is public, and the dormancy recorded in comments.

• https://raw.githubusercontent.com/determined-ai/determined/main/master/internal/license/license.go recorded 2026-09-15

  Determined's own license module: IsEE() returns true when publicKey and licenseKey are both set, and SetLicenseAndKey populates them. Establishes that an enterprise edition is gated on a key.

Adoption is measured by PyPI downloads of the determined package in the trailing 30 days.

• https://pypistats.org/api/packages/determined/recent recorded 2026-09-15

  Trailing-window download counts for the declared PyPI artifact: 62,711 downloads in the trailing month.

Determined's own scheduler allocates slots_per_trial across whole agent machines when the count requires it - setting slots_per_trial to 16 grants two 8-GPU machines together for a single trial - and holds that machine set for the trial. Choosing the machines and holding them together puts Determined level with KAI Scheduler.

• https://raw.githubusercontent.com/determined-ai/determined/main/docs/model-dev-guide/dtrain/dtrain-implement.rst recorded 2026-09-15

  Determined's distributed-training documentation: resources.slots_per_trial enables multi-GPU training, and 'slots_per_trial: 16 # Two 8-GPU agent machines will be used in a trial'.