[Mastra] Document MongoDB differentiators

XMLWordPrintableJSON

    • Type: Task
    • Resolution: Unresolved
    • Priority: Unknown
    • None
    • Affects Version/s: None
    • Component/s: ABX
    • None
    • None
    • None
    • None
    • None
    • None

      Update Mastra's official documentation to reflect the currently implemented but undocumented MongoDB strong differentiators:

       

      Area Gap
      Pre-filter (filterFields) Only a reference param; invisible in concept docs
      VoyageAI reranker VoyageRelevanceScorer/createVoyageReranker ship in code but no doc references them (only Cohere shown)
      BYO positioning Mechanics documented, but "no copy / no sync" value prop unstated (0 hits); absent from RAG overview; no pgvector comparison
      Examples → RAG No MongoDB example (hybrid / BYO / pre-filter)

      Features to document:

      1. Voyage Rerankers

      What it is:
      @mastra/voyageai exports a RelevanceScoreProvider ( VoyageRelevanceScorer / createVoyageReranker ) that plugs into Mastra's rerankWithScorer() and createVectorQueryTool() — but the reranking docs only show CohereRelevanceScorer . Note the Voyage scorer takes a config object (

      { model }

      ), unlike Cohere's string arg.

      a) With rerankWithScorer():

      import { rerankWithScorer as rerank } from '@mastra/rag'
import { VoyageRelevanceScorer } from '@mastra/voyageai'

const scorer = new VoyageRelevanceScorer({ model: 'rerank-2.5' }) // reads VOYAGE_API_KEY

const results = await store.query({ indexName: 'docs', queryVector: embedding, topK: 20 })

const reranked = await rerank({
  results,
  query: 'How do I deploy to production?',
  scorer,
  options: { topK: 5 },
})

      b) With the Vector Query Tool:
       

      import { createVectorQueryTool } from '@mastra/rag'
import { createVoyageReranker } from '@mastra/voyageai'

const tool = createVectorQueryTool({
  vectorStore: store,
  indexName: 'docs',
  model: embedder,
  reranker: { model: createVoyageReranker('rerank-2.5'), options: { topK: 5 } },
})

      2. Vector Search Pre-filters

      Example:

      // 1. Declare filter fields when creating the index
await store.createIndex({
  indexName: 'docs',
  dimension: 1024,
  metric: 'cosine',
  filterFields: ['category', 'tenantId'], // registered as metadata.<field> in the vectorSearch index
})

// 2. Query as usual — filters on declared fields are pushed into $vectorSearch natively
const results = await store.query({
  indexName: 'docs',
  queryVector: embedding,
  topK: 10,
  filter: { category: 'finance', tenantId: 'acme' },
})

// A filter on an UNDECLARED field (or an operator $vectorSearch can't push down)
// automatically falls back to the pre-filter path — no error, just slower.

      

      
            
      

      

          


          

            








            
                        

            Assignee:
            

            
    

    



    
    Unassigned

    
                                            

            Reporter:
            

            
    

            
            



    
            Raschid Jimenez
            
        
    

                                                            
            
        
    
            
    
            
        
                            None
                    
    
            
                                                                                
            Votes:
            
    
            
                    0
        
                                Vote for this issue
            
            
                                
            Watchers:
            
    
            
                    
                    
                    1
        
                            Start watching this issue
            
            
        
            

              









              
            
              
            Created:
    
              

              
    
                                                                                                 

              
            
              
            Updated: