kdeps-components-embedding

version: 1.0.0
requires:
    schemas/v1: ""
    tools/embedding: ""
files:
    component.yaml: |
        apiVersion: kdeps.io/v1
        kind: Component
        metadata:
          name: embedding
          description: Run one embedding operation through the kdeps API
          version: "1.0.0"
        interface:
          inputs:
            - name: operation
              type: string
              required: true
              description: "index, search, upsert, delete, vectorize, embed_query or rerank"
            - name: text
              type: string
              description: Text to index, search for, upsert or delete
            - name: collection
              type: string
              description: Collection (namespace) inside the local embedding database
            - name: dbPath
              type: string
              description: Path to the SQLite embedding database; defaults come from the embedding defaults
            - name: limit
              type: number
              description: Maximum number of search results
            - name: inputs
              type: array
              description: Texts to embed for vectorize and embed_query
            - name: model
              type: string
              description: "Embedding model, for example text-embedding-3-small"
            - name: backend
              type: string
              description: "Embedding backend: openai, ollama, google"
            - name: baseURL
              type: string
              description: Custom base URL for an OpenAI-compatible backend
            - name: rerankQuery
              type: string
              description: Query the candidate documents are ranked against (operation rerank)
            - name: rerankDocuments
              type: array
              description: Candidate documents to rerank (operation rerank)
            - name: rerankTopN
              type: number
              description: Number of top reranked results; 0 returns all
          # Each call posts to the kdeps API tool, which runs the embedding executor;
          # the component output is the executor's structured result.
          output: "{{ fromJSON(output('call').body) }}"
        resources:
          - actionId: call
            name: Embedding operation
            httpClient:
              method: POST
              # In-process host. The HTTP executor calls the handler without dialing.
              url: http://kdeps.api/apis/tools.kdeps.io/v1/tools/embedding/call
              headers:
                Content-Type: application/json
              failOnError: true
              timeout: 120s  # hard stop for the call; vectorize/rerank may hit a provider
              data:
                operation: "{{ get('embedding.operation') }}"
                text: "{{ get('embedding.text') }}"
                collection: "{{ get('embedding.collection') }}"
                dbPath: "{{ get('embedding.dbPath') }}"
                limit: "{{ get('embedding.limit') }}"
                inputs: "{{ get('embedding.inputs') }}"
                model: "{{ get('embedding.model') }}"
                backend: "{{ get('embedding.backend') }}"
                baseURL: "{{ get('embedding.baseURL') }}"
                rerankQuery: "{{ get('embedding.rerankQuery') }}"
                rerankDocuments: "{{ get('embedding.rerankDocuments') }}"
                rerankTopN: "{{ get('embedding.rerankTopN') }}"