API Reference
interfaces
Addautoprocessorconfig

@coasys/ad4m / Exports / AddAutoProcessorConfig

Interface: AddAutoProcessorConfig

Configuration for perspective.addAutoProcessor (everything but uuid).

Table of contents

Properties

Properties

basePrefix

• Optional basePrefix: string

URI namespace new interpreted instances are minted under (the "spawn scope"), e.g. soa://project/42/. Omit for a per-processor default (ad4m://autoprocessor/<processorId>/instance/). Point this and sourceScopeQuery at the same subtree to confine a processor to it.

Defined in

core/src/perspectives/AutoProcessor.ts:162 (opens in a new tab)

batchMax

• batchMax: number

Cap on items per pass.

Defined in

core/src/perspectives/AutoProcessor.ts:178 (opens in a new tab)

batchMin

• Optional batchMin: number

Minimum items before a pass runs (Flux "wait for N inputs"). Default 1.

Defined in

core/src/perspectives/AutoProcessor.ts:176 (opens in a new tab)

claimTtlMs

• claimTtlMs: number

How long a won claim is authoritative before peers may re-claim (ms).

Defined in

core/src/perspectives/AutoProcessor.ts:182 (opens in a new tab)

debounceMs

• debounceMs: number

Quiet-window (ms) after the last new item before a pass runs.

Defined in

core/src/perspectives/AutoProcessor.ts:174 (opens in a new tab)

dedupStrategyJson

• Optional dedupStrategyJson: string

Optional serialized DedupStrategy JSON (else NormalizedString).

Defined in

core/src/perspectives/AutoProcessor.ts:184 (opens in a new tab)

emitDebugEvents

• Optional emitDebugEvents: boolean

Enable full debug observability: persists the raw LLM prompt + response on the pass's InterpretationRun (debugPrompt / debugResponse) AND emits LlmRequestSent / LlmResponseReceived mid-pass events so a subscribed UI can render "waiting on LLM". Default false.

Defined in

core/src/perspectives/AutoProcessor.ts:223 (opens in a new tab)

existingScope

• Optional existingScope: RawScope

Optional parent-scope filter for the dedup lookup: when set, only existing instances of the interpretation classes that live under this scope are candidates for upsert. Omit for the whole-perspective dedup set (pre-scope behaviour). Raw form only — see RawScope.

Defined in

core/src/perspectives/AutoProcessor.ts:199 (opens in a new tab)

flows

• Optional flows: string[]

Canonical flow URIs (SHACLFlow {namespace}{name}Flow) this processor is flow-aware of. Flow features — the flow-aware prompt, the transition-proposal pass, and auto-spawn on freshly created items — run only on the flows listed here, mirroring the class selection above. Omit (or pass []) for no flow processing.

Defined in

core/src/perspectives/AutoProcessor.ts:172 (opens in a new tab)

interpretationClasses

• interpretationClasses: string[]

Class URIs (SHACL target_class) to materialize each pass.

Defined in

core/src/perspectives/AutoProcessor.ts:164 (opens in a new tab)

maxToolCalls

• Optional maxToolCalls: number

Tool-call budget for the interpretation-pass harness this processor runs. Omit / 0 → single-shot LLM path. N > 0 → engage the tool-calling harness with a cap of N calls per pass. Round-tripped through the SDNA via AutoProcessorConfig.maxToolCalls.

Defined in

core/src/perspectives/AutoProcessor.ts:216 (opens in a new tab)

maxWaitMs

• Optional maxWaitMs: number

Safety flush (ms) for a sub-batchMin batch; omit to wait indefinitely.

Defined in

core/src/perspectives/AutoProcessor.ts:180 (opens in a new tab)

mintScope

• Optional mintScope: RawScope

Optional parent-scope target for newly minted instances: when set, every base URI the pass CREATES is additionally linked as a child of mintScope.id via the scope's predicate — turning the SoA-tree "children live under this node" pattern from a URI-prefix convention into an actual graph edge. Upserts of pre-existing instances are NOT linked (would multi-parent unrelated graph state). Raw form only — see RawScope; the Rust watcher rejects Model scopes at runtime.

Defined in

core/src/perspectives/AutoProcessor.ts:209 (opens in a new tab)

processorId

• processorId: string

Human-meaningful processor id (unique per perspective).

Defined in

core/src/perspectives/AutoProcessor.ts:134 (opens in a new tab)

sourceScopeQuery

• sourceScopeQuery: string

SPARQL SELECT ?speaker ?text ?timestamp over the source items to interpret. All three bindings are required: ?timestamp is what makes a turn identifiable, so the processed-turn cursor can tell a re-gathered turn from the same wording said again later. Read the body link's reifier for author and timestamp rather than an app-level ns://author:

SELECT ?speaker ?text ?timestamp WHERE {
  ?m <ns://body> ?text .
  ?r <http://www.w3.org/1999/02/22-rdf-syntax-ns#reifies> <<( ?m <ns://body> ?text )>> .
  ?r <ad4m://ontology/author> ?speaker .
  ?r <ad4m://ontology/timestamp> ?timestamp .
}
ORDER BY ?timestamp

Swap ns://body for your own body predicate. A query binding only speaker+text fails the gather.

Defined in

core/src/perspectives/AutoProcessor.ts:155 (opens in a new tab)

sourceWindowMs

• Optional sourceWindowMs: number

How far back (ms) each pass looks: turns older than now - window are dropped, and the processed-turn cursor only counts runs that finished inside the same window. Omit for no window — every gathered turn is a candidate and the cursor is the unbounded union of this processor's past runs.

Defined in

core/src/perspectives/AutoProcessor.ts:192 (opens in a new tab)

Ad4mmodellikeCalloptions