Inclusion criteria
What gets listed on this site is decided by the written rules below — never by payment, partnership, or persuasion. The rules are deliberately objective so a reader can predict what belongs here and check our consistency against them.
Last reviewed · see also the methodology and the impartiality policy
What qualifies
- Category fit: the product belongs to a tracked category — agent frameworks, orchestration, governance/guardrails, assistants, or the cloud platform layer that powers them.
- Enterprise relevance: the product is plausibly adoptable by a regulated-enterprise team — production posture, governance surface, or a clear path to one.
- Public evidence: official documentation, a product page, or a maintained repository exists and is sufficient to describe the product accurately.
- Provenance: every required schema field can be populated from a resolvable primary source. If the required fields cannot be source-backed, the record waits.
- License clarity: the license can be read from the upstream LICENSE file. Ambiguous or contradictory licensing is disclosed with a warning, not smoothed over.
What stays out
- Research demos, unmaintained experiments, and stealth products without public product evidence.
- Products with no plausible enterprise adoption path (consumer-only tools, hobby projects).
- Coverage-for-coverage's-sake: this is a bounded, opinionated tracker, not an exhaustive AI directory.
Status labels and removal
Being listed is not an endorsement of health — status is tracked honestly and preserved.
- Archived or maintenance-mode projects that remain enterprise-relevant stay listed with an explicit status label — a deprecated tool still deployed in enterprises is exactly what a reader needs to see flagged.
- Records are removed (rather than status-labelled) only when public evidence disappears entirely, provenance re-verification fails permanently, or the product was out of scope to begin with.
- Removals and status changes are dated events in the public change feed, not silent deletions.
Proposing a listing or correction
- Anyone can propose a listing or a correction through the public repository — the proposal must satisfy every rule above and populate the schema's required fields with sources.
- Vendors may propose factual corrections to their own record with sources, exactly like anyone else.
- No fee, sponsorship, badge program, or relationship accelerates or influences listing, placement, or wording — see the impartiality policy.
Where to propose
Use the structured issue templates in the repository (Propose a tool, Data correction, License correction) or open a PR per the CONTRIBUTING guide — each requires the primary sources that back every field. CI validates schema, provenance, and license gates automatically before human review; proposals that fail are declined with the failing rule named.