Trust, with context.

Understand what the platform does, where information goes and what still needs review.

Follow the source.

Model metadata and benchmark claims link back to their publishers. A public repository or permissive weight license does not establish suitability for every use case.

Keep your evidence.

Export model revisions, evaluation context, assumptions and review notes in a portable pack. Your team can inspect the record outside AltOpenAI.

Make an accountable choice.

The workbench checks documented evidence against your criteria. It does not grant approval, certify compliance or establish model safety.

CAPABILITY REGISTER

Available today. Planned next.

Reviewed 20 September 2026

CapabilityScopeAvailability

Source and revision

Capture publisher metadata with an exact model revision, source link and retrieval time. Publisher claims remain attributed.

Available

Portable decision packs

Save a draft in this browser and export it as JSON. Imports, measured results and review notes remain user-supplied evidence.

Available

Evaluation context

Check supplied results against the entered benchmark context, freshness and acceptance thresholds. Evaluation execution and independent verification are not provided.

Available

Team identity and access

Shared accounts, SSO, SCIM, permissions and authenticated approval need a provisioned identity and tenant data service.

Planned

Audit, retention and residency

Immutable audit history, cloud retention policies, residency commitments and contractual service levels require operational controls and verification.

Planned

Private hosting and compute

Private repositories, weight uploads, managed evaluations and GPU endpoints need storage, isolated workers, quotas and metering.

Planned

DATA FLOW

Where your information goes.

Discovery: searches and repository requests pass through AltOpenAI to public upstream services. The hosting infrastructure processes these requests.
Local drafts: saved models and workbench packs use browser storage. Workbench imports are read locally; they are not uploaded to a shared workspace.
Playground: your Hugging Face token stays in browser memory. Prompts and the token are sent directly to the Hugging Face router for provider inference.
Exports: downloaded JSON contains your entered notes and evidence. Review its contents before sharing; it is an unsigned working document.

Browser storage is not a private enterprise repository. Other people with access to this browser profile may access drafts; clearing browser data can remove them. Keep an exported copy and avoid adding confidential source data.

Understand Hugging Face provider routing and billing ↗

DESIGN REFERENCES

Standards inform the workflow.

NIST AI Risk Management Framework ↗Use-case context, risk ownership, measurement and ongoing review.
OWASP ASVS 5.0 ↗A verifiable application-security baseline for future shared services.
OWASP Top 10 for LLM Applications ↗Prompt injection, supply-chain risk and excessive agency review prompts.
SLSA provenance ↗Artifact origin, production context and supply-chain traceability.

These are design references, not AltOpenAI certifications. SOC 2, ISO certification, contractual SLAs and independent assurance are not offered in this release.