Skip to main content

RAG Module

The rag module indexes sources into knowledge bases and retrieves for agents. This page matches internal/modules/rag in the Go daemon. Older docs that listed fourteen agent tools (including multi_query, sql_query, ingest_database, clear_cache, list_models) were wrong for this build.

Authoritative tool table: reference/modules/rag.

Agent tools (11)​

create_knowledge_base, delete_knowledge_base, list_knowledge_bases, knowledge_base_stats, index_stats, ingest, ingest_file, ingest_directory, query, reindex, migrate_embeddings.

There is no separate vector tool module in digitornd.

Zero-config​

yaml
tools:
  modules:
    rag: {}

Configuration (high level)​

Config is bound from YAML under tools.modules.rag (and related module config). Important groups from Config in Go:

AreaNotes
embedding_modelString shortcut or {id, dimensions, pooling}
backendVector store settings (Qdrant / pgvector / Elasticsearch appear in code)
pipeline / chunking / citations / cache / aclRetrieval and safety knobs
sourcesFile, DB, web, kafka-style source entries for the indexer
auto_indexon_start, schedule (cron string for indexer triggers, not cron_native)
default_knowledge_baseDefault KB name
max_knowledge_bases / max_documentsCaps

Source entries can carry their own triggers (type, every, cron). Exact field validation lives in the module and its tests; do not invent agent tools for Text2SQL or multi-query if they are not in the RegisterTool list above.

Example​

yaml
tools:
  modules:
    rag:
      config:
        embedding_model: minilm-l12
        default_knowledge_base: docs
        auto_index:
          on_start: true
          schedule: ""   # optional cron for indexer; not cron_native
        sources:
          - name: handbook
            type: file
            path: "{{workdir}}/docs"
            extensions: [.md, .txt]
            recursive: true
  capabilities:
    grant:
      - module: rag