Knowledge Extraction scaffold

A complete program, not a fragment: it compiles as written. Copy it, rename the placeholder identifiers to your domain, and run axon check — the compiler will tell you what your renaming broke. An MCP client can also generate it through the axon.compose tool by naming the knowledge_extraction domain.

// AXON Knowledge Extraction scaffold — entity + relationship
// extraction for knowledge-graph construction.
//
// Differs from `document_analysis.axon` (general bulk processing):
// this scaffold targets structured KG construction — typed entity
// candidates + typed relationship triples + confidence per claim.

// ── Types ─────────────────────────────────────────────────────────

type RawText {
    source_id:    Text
    content:      Text
    captured_at:  Text
}

type EntityCandidate {
    surface_form:  Text
    canonical_uri: Text
    entity_type:   Text
    confidence:    Number
}

type RelationTriple {
    subject_uri:   Text
    predicate:     Text
    object_uri:    Text
    confidence:    Number
    evidence_span: Text
}

type ExtractionResult {
    source_id:   Text
    entities:    Text
    relations:   Text
    summary:     Text
}

type ExtractRequest { text: RawText }

// ── Persistent KG store ──────────────────────────────────────────

axonstore KnowledgeGraph {
    backend:    postgresql
    connection: "postgres://kg.internal/graph"
    isolation:  serializable
    on_breach:  raise
    schema {
        triple_id:    Uuid primary_key
        subject_uri:  Text not_null
        predicate:    Text not_null
        object_uri:   Text not_null
        confidence:   Numeric not_null
        evidence:     Text
        recorded_at:  Timestamp not_null
    }
}

// ── Identity + grounding ──────────────────────────────────────────

persona KGExtractor {
    domain: ["knowledge-graphs", "information-extraction", "ontology-design"]
    tone: precise
    confidence_threshold: 0.85
    cite_sources: true
}

context KGExtractionContext {
    memory: session
    language: "en"
    depth: deep
    max_tokens: 4096
    temperature: 0.1
}

anchor EvidenceSpanRequired {
    require: source_citation
    confidence_floor: 0.8
    unknown_response: "Evidence span insufficient — skipping triple."
    on_violation: log
}

// ── Shield ────────────────────────────────────────────────────────

shield KGShield {
    scan:       [prompt_injection, pii_leak, hallucination]
    on_breach:  quarantine
    severity:   high
    compliance: [SOC2]
}

// ── Flow ──────────────────────────────────────────────────────────

flow ExtractKnowledge(text: RawText) -> FlowEnvelope<ExtractionResult> {
    step IdentifyEntities {
        given: text
        ask: "Identify entity candidates; emit canonical URIs where confident; cite span ranges."
        output: FlowEnvelope<ExtractionResult>
    }
    step RelateEntities {
        given: IdentifyEntities.output
        ask: "Derive relation triples; cite evidence span per triple; emit confidence."
        output: FlowEnvelope<ExtractionResult>
    }
    return RelateEntities.output
}

// ── HTTP boundary ─────────────────────────────────────────────────

axonendpoint ExtractionAPI {
    method:      post
    path:        "/v1/kg/extract"
    body:        ExtractRequest
    execute:     ExtractKnowledge
    output:      FlowEnvelope<ExtractionResult>
    shield:      KGShield
    backend:     auto
    compliance:  [SOC2]
    retries:     1
    timeout:     60s
    requires:    ["kg.extract"]
}