Skip to content

TODOS.md

Latest commit

 

History

History
35 lines (28 loc) · 2.08 KB

TODOS.md

File metadata and controls

35 lines (28 loc) · 2.08 KB

Cosmos OSS — Deferred Work

P1 — High Priority

SQLite connection pool (r2d2_sqlite)

  • What: Replace single Arc<Mutex<Connection>> in database_service.rs with a proper connection pool. Each worker gets its own connection, eliminating mutex contention.
  • Why: Root cause of DB corruption under heavy indexing load. The hotfix adds busy_timeout and poison recovery as a band-aid, but the real fix is eliminating the single-mutex bottleneck.
  • Blocked by: r2d2_sqlite v0.33 requires rusqlite v0.39, but project uses v0.37. Need to upgrade rusqlite first, which may require sqlite-vec compatibility check.
  • Effort: M (human) / S (CC). Touches database_service.rs + every service that calls get_connection().
  • Added: 2026-04-05 via /plan-eng-review (performance hotfix)

P2 — Next Major Features

Replace Whisper with Gemma 4 for transcription

  • What: Once Gemma 4 (or future version) supports timestamped audio segments, replace Whisper entirely. Reduces model count from 3 to 2.
  • Why: Simplifies model stack. One fewer runtime dependency (Candle Whisper pipeline).
  • Blocked by: Gemma 4 lacking timestamp/segment output. TranscriptionDisplay.tsx depends on segmented transcript data.
  • Effort: M (human) / S (CC)
  • Added: 2026-04-05 via /plan-ceo-review

RAG interface — "Ask questions about your files"

  • What: Chat-like interface for natural language Q&A grounded in indexed files. Gemma 4 E2B + Nomic retrieval = local RAG.
  • Why: The ultimate personal knowledge base feature.
  • Depends on: Gemma 4 integration, clustering, and search-within-cluster all working.
  • Effort: L (human) / M (CC)
  • Added: 2026-04-05 via /plan-ceo-review

P3 — Future

Model hot-swap for custom GGUF models

  • What: Let users/contributors swap the Gemma 4 GGUF for any compatible model (Llama, Phi, Mistral) via config.
  • Why: OSS extensibility. Contributors can experiment with different models.
  • Depends on: llama-cpp-rs integration being stable.
  • Effort: S (human) / S (CC)
  • Added: 2026-04-05 via /plan-ceo-review