Semantic search fails: No embeddings available despite enabled status (v0.9.13)

[madeinoz67]

Description

Semantic search fails with "No embeddings available for search" error despite index containing symbols and semantic search being enabled.

Environment

• codanna version: 0.9.13 (latest)
• OS: macOS (Darwin 25.0.0)
• Project: TypeScript codebase

Reproduction

1. Index a codebase:

codanna index src/

2. Attempt semantic search:

codanna mcp semantic_search_with_context query:"entity registry file path creation entities.json" limit:5

Error Output

Semantic search failed: Semantic search error: No embeddings available for search

Index Status

Running codanna mcp get_index_info shows:

Index contains 1052 symbols across 31 files.

Breakdown:
  - Symbols: 1052
  - Relationships: 733

Symbol Kinds:
  - Classs: 15
  - Constants: 585
  - Enums: 1
  - Fields: 57
  - Functions: 33
  - Interfaces: 73
  - Methods: 234
  - TypeAliass: 11
  - Variables: 43

Semantic Search:
  - Status: Enabled
  - Model: AllMiniLML6V2
  - Embeddings: 0          <--- THIS IS THE PROBLEM
  - Dimensions: 384
  - Created: 10 hours ago
  - Updated: 10 hours ago

Expected Behavior

Semantic search should work when index contains symbols and semantic search is enabled.

Notes

• Regular symbol search works fine (search_symbols, find_symbol)
• Index shows "Status: Enabled" but "Embeddings: 0"
• Similar to closed issue Semantic search failed: Semantic search failed: No embeddings available for search #41, but still occurring in latest version (0.9.13)

[madeinoz67]

Workaround Found

Root Cause: Regular codanna index does not generate embeddings when an index already exists. The embeddings are only created during initial indexing or when forcing a full reindex.

Workaround: Use the --force flag to regenerate embeddings:

codanna index --force src/

Evidence

Without --force:

EMBED: [                    ]   0%  0/262 embedded | 0/s

Embeddings stay at 0.

With --force:

EMBED: [▮▮▮▮▮▮▮▮▮▮▮▮▮▮▮▮▮▮▮▮] 100%  262/262 embedded | 36/s

Result:

Semantic Search:
  - Status: Enabled
  - Model: AllMiniLML6V2
  - Embeddings: 262 (was 0)
  - Dimensions: 384

Semantic search now works correctly and returns relevant results with proper similarity scores.

Expected Behavior

When semantic_search.enabled = true in settings.toml, incremental index updates should maintain embeddings for new/modified symbols, not just during initial or forced reindexing.