System Layer — Shared Services Behind HEPO
Understand HEPO's shared security, data, integration, local AI, workspace, and extension services, and how to evaluate them in your deployment.
Local AI: inference within your deployment
Understand how local inference supports HEPO evaluation, retrieval, and assistant capabilities, with capacity and access configured for the intended use.
Understand how local inference supports HEPO evaluation, retrieval, and assistant capabilities, with capacity and access configured for the intended use.
Local AI is HEPO's approach to running model inference within the organisation's deployment. The platform's evaluation, retrieval, and assistant capabilities use the configured local service where their operation requires model interpretation. The organisation and implementer establish the endpoint, model, capacity, and access around that use.
This keeps the intended inference path inside the operating boundary. The model is a shared service for the offerings, rather than another sellable offering or an authority that replaces the responsible person.
Follow a question through the local service
A user asks the assistant a question about permitted work. The supported retrieval and tool paths supply relevant context, the configured local model interprets the request, and the application presents an answer or supported action. The usefulness of that response depends on the available records, retrieved context, and model behaviour.
For an evaluation task, the configured pipeline may use AI-assisted observation to interpret relevant evidence against an organisation-owned definition. The reading still needs its evidence and context; a model's interpretation is not an independent fact about a person or process.
Where the capability is used
| Consumer | Role of local inference |
|---|---|
| Evidence interpretation | Assist with semantic observations where the configured evaluation path uses them |
| Knowledge retrieval | Interpret a question and use relevant supplied context |
| HEPO Assistant | Support conversation, navigation, and supported tool or process requests |
| Connected-record interpretation | Use available relationships and record context in an answer or reading |
Relationships supplies meaningful connections; Content supplies knowledge. Their existence does not mean the model can access every record or automatically understand every relationship.
Define and verify the inference boundary
HEPO's deployment policy keeps AI inference local and does not require sending tenant data to an external model service. The implementation needs to configure the actual inference endpoint and the network, credentials, and logging appropriate to that policy.
External business integrations remain separate exchanges with their own scope. Local inference does not imply that an authorised connector never transmits information to another business system. The Deployment page places both kinds of connection in the overall boundary.
Plan for useful operation
Model choice, context size, concurrent use, and hardware capacity affect latency and the work the service can handle. Use representative questions and evidence when sizing the deployment. Examine unavailable-service behaviour as well as a successful response.
Business owners define the questions and acceptable review process. Implementers configure and operate the local service. Reviewers check representative answers, the sources supplied to them, and the effect of restricted access. Human decisions and approval gates remain explicit in Process Orchestration and People & Culture Orchestration.