docs: add semantic search implementation plan #110
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Plan for implementing semantic search with local embeddings: - LanceDB for vector storage (embedded, serverless) - FastEmbed-rs for local embedding generation (ONNX-based) - BGESmallENV15 as recommended embedding model - Hybrid search combining semantic + existing FTS5 - 5-phase implementation roadmap
- Add timestamp range filtering (started_at, created_at) - Add provider, project, outcome filters for session search - Add turn_type filter for turn search - Document LanceDB scalar indexes for filter performance - Add SQL filter syntax examples - Update CLI commands with filter options
- Document LanceDB as fully embedded (no external server) - Detail ONNX Runtime linking options (static/dynamic/download) - Document model file strategies (runtime download/local/embedded) - Recommend runtime download for balance of size vs convenience - Add alternative compile-time embedding for offline deployments - Update risks table with ONNX linking mitigation
- Add to_embedding_text() implementations for TurnSummary and SessionSummary - Include all relevant fields: user_intent, assistant_action, summary, key_topics, decisions_made, code_concepts - SessionSummary embedding includes ChatSession context (provider, project) - Add TurnEmbedding and SessionEmbedding data structures - Document complete indexing flow with change detection via SHA256 hash - Add example output for both embedding text formats
Implement Phase 1 of semantic search for turn and session summaries: - Add embedding/ module with EmbeddingService wrapping FastEmbed-rs - Support for multiple models (BGESmallENV15, AllMiniLML6V2, etc.) - Configurable cache directory and download progress - Batch embedding generation for efficiency - Add vector_store/ module with VectorStore wrapping LanceDB - Arrow schemas for turn and session embeddings - Upsert, search, and delete operations - Metadata filtering via SQL WHERE clauses (date ranges, types, providers) - Cosine similarity search with configurable limits - Add to_embedding_text() and embedding_text_hash() to TurnSummary - Combines intent, action, summary, type, topics, decisions, concepts - SHA256 hash for change detection to skip unchanged re-embedding - Add to_embedding_text() and embedding_text_hash() to SessionSummary - Combines title, summary, goal, outcome, provider, project, entities - Context-aware hash including provider and project - Feature-gate all semantic search code behind "semantic-search" feature - Add workspace dependencies: lancedb, arrow, fastembed, sha2
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Plan for implementing semantic search with local embeddings: