EdgeData API reference

EdgeData provides local data collection and classification primitives for Edge apps.

It is supporting infrastructure. Keep app-specific business rules in the app.

Edge

public enum Edge

Static API for local data operations.

Method	Description
`bootstrap(dbQueue:trainingDataSink:)`	Initializes EdgeData with a local GRDB queue and optional training sink.
`registerSchema(_:)`	Registers an app data shape at launch.
`schema(_:)`	Looks up a registered data shape.
`registeredSchemaNames()`	Lists registered names.
`recordRaw(fact:customFactId:)`	Records unclassified local data.
`record(fact:)`	Records already-classified local data.
`recordEvent(...)`	Records an AI interaction event.
`countFeedbackEvents(namespace:)`	Counts events with feedback.
`queryFacts(namespace:schema:status:limit:)`	Queries local facts.
`countFacts(namespace:schema:status:)`	Counts local facts.
`correctClassification(...)`	Applies a user correction to a classification.
`onClassified(_:)`	Registers a callback for successful classification.
`onClassificationFailed(_:)`	Registers a callback for failed classification.
`fetchFact(id:)`	Fetches one fact by ID.
`reclassifyAsync(...)`	Requests reclassification.

Type	Description
`Sensitivity`	`.localOnly`, `.meshOk`, `.trainingOk`.
`FactStatus`	Local classification lifecycle state.
`FieldType`	Supported field kinds for app data shapes.
`FieldDef`	Field declaration.
`SemanticLabels`	Labels for primary timestamp, value, and entity fields.
`SchemaDef`	App data shape declaration.
`RawFact`	Raw local payload waiting for classification.
`Fact`	Classified local payload.
`QueryStatus`	Query filter for classified, pending, failed, or all facts.

public actor ClassificationDaemon

Background classifier for raw local data.

API	Description
`ClassificationDaemon.shared`	Shared daemon.
`start(namespace:candidateSchemas:llmClient:toolNames:promptBuilder:)`	Starts classification for a namespace.
`stop()`	Stops the daemon.
`wake()`	Wakes the daemon from idle sleep.

public protocol EdgeClassificationLLMClient: Sendable

The app implements this protocol to provide local LLM classification.

Requirement	Description
`isBusy`	Whether the local LLM is busy with user-facing work.
`generate(messages:)`	Generates a classification response.
`generate(messages:toolNames:)`	Optional tool-aware generation path.

public enum ClassificationParser

Parses local LLM classification output.

Method	Description
`parse(llmOutput:candidateSchemas:)`	Parses and validates a classification response.
`stripMarkdownCodeFence(_:)`	Removes a Markdown code fence wrapper.
`extractJSONObject(_:)`	Extracts the first JSON object from text.

Use Sensitivity to keep local-only data local. Do not include raw user correction text in logs, analytics, or collaboration messages.