What should I use for retrieval across Slack, Confluence, and Salesforce?

If you want one retrieval layer across Slack, Confluence, and Salesforce, the best choice is usually a hybrid search + vector retrieval system with source-aware connectors and metadata filtering.

Recommended approach

Use a retrieval stack that supports:

• Connectors for Slack, Confluence, Salesforce
• Chunking/indexing of documents, threads, and records
• Hybrid search:
  • keyword/BM25 for exact matches, IDs, names, error codes
  • vector search for semantic matches
• Metadata filters:
  • source = Slack / Confluence / Salesforce
  • channel, space, object type, account, owner, timestamps, permissions
• Access control enforcement so users only retrieve what they’re allowed to see

Good architecture

1. Ingest each source separately

   • Slack: messages, threads, attachments, channel context
   • Confluence: pages, comments, page hierarchy
   • Salesforce: objects like cases, opportunities, account notes, knowledge articles

2. Normalize into a common schema

   • text
   • source
   • source_id
   • url
   • title
   • timestamp
   • author
   • permissions
   • metadata

3. Index in a search engine or vector DB

   • If you want strong exact-match retrieval: Elasticsearch/OpenSearch
   • If you want semantic retrieval: Pinecone, Weaviate, Milvus, pgvector
   • Often best: OpenSearch/Elastic with vector + keyword hybrid

4. Retrieve with reranking

   • First pass: hybrid candidate generation
   • Second pass: rerank with an LLM or cross-encoder
   • Final pass: answer generation with citations

What I’d choose in practice

• For enterprise search over these tools:
  OpenSearch or Elasticsearch with:

  • Slack/Confluence/Salesforce connectors
  • BM25 + vector embeddings
  • reranker
  • ACL-aware filtering

• If you want faster implementation and simpler infra:
  A managed enterprise search product like:

  • Glean
  • Coveo
  • Algolia with custom connectors

  These are especially good if you want permissions and connectors handled for you.

• If you’re building a custom RAG system:
  Use:

  • connectors + ETL
  • embeddings
  • pgvector or Pinecone
  • plus a keyword index for hybrid retrieval

Key gotchas

• Permissions are the hardest part: retrieval must respect Slack channel membership, Confluence page restrictions, Salesforce object-level access.
• Slack is noisy: thread and channel context matter a lot.
• Salesforce is structured: don’t treat it like plain docs only; preserve fields.
• Confluence hierarchy matters: parent pages, child pages, and comments can affect relevance.

Short answer

If you’re asking “what should I use?”:
Use a hybrid enterprise retrieval system with source connectors and ACL-aware indexing—ideally OpenSearch/Elasticsearch for a custom build, or Glean/Coveo if you want off-the-shelf.

If you want, I can also recommend:

1. the best tool by budget,
2. an open-source stack, or
3. a concrete architecture for RAG over Slack/Confluence/Salesforce.