Current evidence
Prototype fallback route
Hermes has a prototype model-routing fallback path that can name an Anthropic/Claude route. That is an architectural prototype and route definition, not proof that Claude is being called for customer work.
API readiness · evidence as of 2026-10-09
A bounded review path, not a claimed live integration.
AI Clouds Company OS is exploring Claude as a focused quality reviewer for generated website content. The current evidence supports a prototype model-routing fallback in Hermes; it does not verify direct Anthropic SDK/API use in a customer-facing product.
Not verified
No direct live Anthropic API claim
We have not independently verified a live direct Anthropic SDK/API invocation, a customer-facing Claude call, or customer value produced by Claude. Developer tooling and a configured fallback route are not substitutes for a product receipt.
How to read this page: “Proposed” describes future evaluation work. “Verified” is reserved for evidence tied to the current source, runtime and an accepted review receipt. This page publishes neither credentials nor internal endpoints.
Proposed next milestone
A bounded quality reviewer for generated sites
The intended Claude integration is narrow: inspect a generated website package and return review findings for a human decision-maker. It is not an autonomous publisher, business-fact authority or replacement for source evidence.
Detect
Flag ungrounded site facts, unsupported claims, omissions and content-quality issues against the supplied evidence package.
Evaluate
Measure reviewer performance on a human-reviewed evaluation dataset, including both expected findings and safe abstentions.
Gate
Keep business-fact corrections, irreversible actions and publication behind explicit human review and approval gates.
Evidence required before a stronger claim
What a responsible pilot would need to show
- A current-source, current-runtime receipt that identifies the intended Claude call and its bounded review task.
- A human-reviewed evaluation set with documented findings, misses, false positives and abstentions.
- Independent checks that confirm factual grounding, safe handling of uncertainty and no automatic publication.
- Human approval for business facts, customer commitments, production changes and other consequential effects.
Until those checks exist, the accurate status remains prototype route / direct API use unverified. Internal previews and engineering artifacts do not establish live Claude customer usage.
Funding and technical support are hypothetical
If available, Anthropic credits or technical support could help us test this bounded reviewer, expand the evaluation set and prepare a responsible pilot. This is a funding/support hypothesis, not a promise of eligibility, acceptance, credits or support.
Review the surrounding evidence
See the dated engineering snapshot and the governance concepts that keep proposed model use bounded.