multi_backend.mnm

# multi_backend.mnm
# Multi-backend memory system with aliased query scoring

config {
  storage: sqlite("./data/multi.db")
  vector: {
    primary: auto("qdrant"),
    fast: auto("chromadb")
  }
  graph: {
    kg: auto("neo4j"),
    social: auto("memgraph")
  }
  embedding: openai("text-embedding-3-small")
  extractor: openai("gpt-4o-mini")
}

namespace multi_agent {

  memory Knowledge {
    content: string
    subject: string
    predicate: string
    object: string
    confidence: float[0,1]
    source: string?
    created_at: timestamp

    @index subject
    @index predicate
  }

  on ingest(content: string) {
    content
      |> extract()
      |> store()
  }

  # Query using aliased scores for custom ranking
  query Search(input: string): Knowledge[] {
    from Knowledge
    where semantic_match(input, backend: "primary", threshold: 0.5) as sm
      and graph_match(input, backend: "kg", hops: 2) as gm
    order by sm * 0.6 + gm * 0.4 desc
    limit 20
  }

  # Query with pure semantic match on a specific backend
  query FastSearch(input: string): Knowledge[] {
    from Knowledge
    where semantic_match(input, backend: "fast", threshold: 0.7) as score
    order by score desc
    limit 10
  }

  update Knowledge {
    on_conflict(old, new) {
      if new.confidence > old.confidence {
        supersede(old, new)
      } else {
        discard(new)
      }
    }
  }
}