Skip to content

AI-powered semantic search via Meilisearch embeddings + RAG #367

@sanxiago

Description

@sanxiago

Is your feature request related to a problem? Please describe.

When searching archived emails, results are limited to exact keyword matches. This makes it difficult to find emails by concept or intent searching "budget approval" won't surface "sign-off on Q3 spend" even though they're semantically equivalent. For e-Discovery and compliance use cases, this is a significant gap since relevant emails often use varied or indirect terminology.


Describe the solution you'd like

Enable semantic (vector) search and RAG (Retrieval-Augmented Generation) over archived emails by leveraging Meilisearch's built-in embedding and hybrid search support, which is already available since Meilisearch v1.6.

Since OpenArchiver already uses Meilisearch as its search backend, no additional vector database or search infrastructure is required. Meilisearch can be configured with an embedder (e.g. Ollama, OpenAI, or HuggingFace) to automatically generate and store vector embeddings at index time, and its native hybrid search combines full-text and semantic ranking in a single query.

At a high level, this would allow users to:

  • Find emails by meaning, not just keywords
  • Ask natural language questions against their email archive and get cited, grounded answers

Describe alternatives you've considered

  • External vector database (Qdrant, Chroma, pgvector) adds infrastructure complexity and duplicates data already in Meilisearch. Unnecessary given Meilisearch's native capabilities.
  • Meilisearch's built-in /chat endpoint viable for a basic implementation, though a custom RAG layer would allow better control over scoping, citation tracking, and conversation history.

Additional context

Reference: How to build a RAG system (with Meilisearch)

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions