How Frontal works

Frontal connects the layers an intelligent system needs instead of making each application assemble them independently.

CopySource systems
    ↓
Connectors, syncs, and data quality
    ↓
Ontology: entities, relationships, and permissions
    ↓
Context: the facts relevant to one request or run
    ↓
Agents and workflows
    ↓
Runs, actions, and artifacts
    ↓
Evaluation, observability, security, and governance

1. Bring in trusted context

Connectors and data pipelines acquire records from enterprise systems. The data layer keeps provenance, schema information, validation signals, and freshness close to those records. This makes a source useful as evidence, not merely as text for retrieval.

2. Give the data shared meaning

The ontology defines what the business deals in: entities such as accounts, invoices, requests, and policies; their properties; and their relationships. Source systems map into this model so a developer does not need to teach each agent a different schema.

3. Execute work inside boundaries

An agent uses context and tools to choose a path toward an objective. A workflow follows an explicit sequence of steps and can call an agent where judgement is useful. Both execute as runs with an identity, limits, policy checks, and a recorded outcome.

4. Learn from every run

Evaluation tests whether a proposed version meets known requirements. Observability explains a specific production run. Governance controls access, tool use, approvals, and audit. Together they make improvement deliberate rather than anecdotal.

One system, not a feature stack

The layers are designed to share boundaries. The same entity model can serve an API, an agent, a workflow, and a user interface. The same permission decision can constrain a person, an integration, and a tool call. The same run trace can support debugging, evaluation, and audit.

Next: Context and ontology.