About & coverage

The Compute Terms Observatory compiles, in one continuously updated and source-linked place, what cloud infrastructure and AI model providers actually publish in their terms. Each provider's documents are archived twice daily, read against a fixed schema of contract dimensions, and laid out side by side.

Assembled this way, the terms support three things that are awkward to do by reading provider sites one at a time: seeing market posture across providers on the same dimension, noticing when a provider's terms change, and reaching the primary document behind any value in one click.

Audiences who have found this kind of assembly useful include counsel preparing negotiations, researchers, journalists, and procurement teams. Whether it suits any particular purpose is a judgment only you can make, and every value links to its source so you can check it.

It describes what the documents say. It never advises what to do.

Public documents only: terms of service, SLAs, acceptable-use and usage policies, model licenses, and deprecation policies for 27 providers across 29 contract dimensions. Each value carries a short verbatim quote and a link to the archived source. See the methodology for the full pipeline and coverage.

An automated workflow re-fetches every tracked document twice daily and archives a timestamped, content-hashed snapshot. Each section header shows when its corpus was last checked.

No, verify before you rely on it. These are an AI's readings of public terms, not the terms themselves and not legal advice. They can be wrong, incomplete, or out of date, and public terms are only a starting point; negotiated agreements routinely differ. Every value links to its source so you can check it yourself. Nothing here creates an attorney-client relationship.

It is an independent research project. The source code is published so the methods can be inspected and the published claims checked, but it is source-available rather than open source: it is not licensed for reuse. The brand is reserved. It is not affiliated with, endorsed by, or sponsored by any of the providers it tracks.

Open an issue on the source repository with the provider, the dimension, and what you believe is wrong. Corrections are re-checked against the archived source.

Errors fall into three kinds, and it helps to say which:

• The reading is wrong. The cited document says something different from what the cell reports.
• The document is out of date. The provider has since revised or replaced it.
• The cited document does not govern the tracked artifact. The document is real and quoted correctly, but it does not apply to the thing in that row: a hosted platform's terms cited against downloadable weights, a policy that governs a different product of the same company, or a document published by someone other than the provider. Every document carries a stated basis for why it governs; this is a challenge to that basis.