How HiveAI providers and models are governed¶
An AI provider runs the service that produces an answer or prepares information for search. A model is the particular deployment used for that work. Zerohive selects approved configurations on the server so that a member's question cannot silently move information to an arbitrary provider or region.
For executives and business owners, provider administration is a change-control responsibility: who may process the information, for which purpose, in which region, under which terms and spending limits, with what evidence and recovery plan. A model's quality alone does not answer those questions.
Availability: non-private HiveAI capabilities are available by default subject to licensing, permissions, governance and credits. Each environment still needs its approved provider resources and working configuration. The frontend is in progress; this draft explains implemented boundaries and the intended governance experience. It does not establish a deployed model picker or provider-settings form.
What the different decisions mean¶
| Decision | What it establishes | What it does not establish |
|---|---|---|
| A capability is included in the plan | Commercial eligibility | A user's permission or available provider |
| A provider has current approval | Permission to use it within the approved purpose and boundaries | A working endpoint or permission to read every source |
| A model configuration is enabled | The server may consider that configuration | Compatibility with every requested capability |
| Configuration checks pass | Required settings and selected constraints are present | A successful provider connection or answer |
| The current action is authorized | The user and operation meet current access rules | A final charge or completed business action |
| A response is completed | The relevant workflow reached its recorded outcome | Guaranteed factual accuracy or approval of a subsequent action |
An unavailable configuration stays unavailable. The inspected tenant resolver requires one exact approved provider, profile and model registration. An alternative model is a deliberate governed change, not an automatic way to bypass a failed check.
Responsibilities¶
Members use the capabilities made available to them and report problems with safe references. Workspace administrators review access, plans, budgets and the current availability explanation. They should not change a provider endpoint or credentials in order to make a missing feature appear.
Platform governance owners maintain policies, supplier evidence, oversight and incident decisions. Authorized operators manage the approved configuration and delivery. Configuration changes and governance decisions are related but separate: recording supplier evidence does not install a model, and changing a setting does not approve a supplier.
The tenant assistant, the platform staff assistant and knowledge processing have distinct runtime owners. A successful staff-console response does not demonstrate that tenant conversations or manual processing are ready. Staff assistance also retains its separate access and accounting boundary.
Changes can affect existing work¶
A provider, model, region, privacy term, system instruction, tool behavior, retrieval setup or usage-accounting change can alter the meaning or safety of an operation. Preserve the earlier policy and configuration evidence. Review the effect on pending conversations, approved actions and paid knowledge processing before rollout or rollback.
Changing an embedding model can affect the search representation of existing documents. A successful chat test does not prove that stored knowledge can still be retrieved correctly. See the knowledge lifecycle guide.
Use the right guide¶
Use the provider availability and change guide for user recovery, administrator responsibilities and the business review sequence. The governance guide explains policy and incident decisions; the credits guide explains reservations and charges. Private Cloud remains disabled in the inspected configuration and is not an automatic fallback.
Private deployment assessment¶
The Private Cloud boundary remains disabled and has its own preparation and support responsibilities. A provider change, commercial contract or healthy configuration check does not establish a qualified private deployment.