Editorial methodology

We prefer evidence in this order:

1. Official documentation for current setup, supported behavior, architecture, and operating guidance.
2. The official source repository for code, license, releases, changelog history, and issue context.
3. Direct first-party product surfaces for currently displayed plans, terms, and interfaces, with the observation date kept close to time-sensitive claims.
4. Reproducible first-party testing for workflows we can run and describe with the relevant configuration and limitations.
5. Third-party sources only when they are necessary to describe an external product, market, or independent viewpoint. They do not override official Hermes Agent sources.

Search snippets, copied feature lists, and unsourced summaries are discovery aids, not final evidence.

Claims are checked against the most relevant available source before publication. Configuration-dependent behavior is described with its dependency instead of being presented as universal.

Pages with volatile details such as pricing, provider support, release behavior, or commercial terms should show an updated date. A date is evidence of the last editorial pass, not a guarantee that every external surface stayed unchanged afterward. Readers should use the linked official source for live operational decisions.

Reviews separate observed facts from editorial judgment. We identify what was inspected or tested, the relevant environment or plan when it changes the result, and material limitations. Comparisons use the same decision criteria across products and distinguish open-source software, self-hosted operation, and managed services.

We do not publish aggregate ratings, invented reviewer identities, or “best” claims as structured data without support. A commercial relationship does not convert a marketing statement into evidence.

Product claims must be traceable to an official source or a reproducible observation. Examples and hypothetical workflows are labeled as examples; they are not presented as measured outcomes. Case studies should identify the subject, method, timeframe, and evidence that can be disclosed.

Anonymous anecdotes, generated testimonials, unverifiable performance numbers, and unsupported experience claims are not treated as proof. Where evidence is incomplete, the page should narrow the claim or say what remains unverified.

Material errors are corrected in place after checking the underlying evidence. A correction should update affected claims and links, preserve clear qualification where uncertainty remains, and refresh the page date when the change is substantive. We do not silently turn an unverified claim into a stronger one.

To report an editorial error, open an issue in the website repository ↗ with the page URL, disputed text, and a primary source when available. Product bugs belong in the official Hermes Agent issue tracker ↗.

This site can link to commercial services, including managed Hermes Agent offerings, and can publish affiliate-oriented content. A page containing an affiliate or other compensated link should disclose that relationship near the relevant recommendation. Commercial value does not change the source hierarchy or verification standard.

Hermes Agent’s MIT-licensed source code, the infrastructure and model services needed to run it, and separately sold managed services are distinct. We do not describe the open-source repository as though it were a paid subscription, or imply that a commercial service is included with self-hosting when it is not.