LearnNewsExamplesServices
Frontmatter
id9710
title[SQLite VSS Migration] 100% Offline Markdown Tensor Chunking
stateClosed
labels
enhancementaiarchitecture
assigneestobiu
createdAtApr 4, 2026, 11:05 PM
updatedAtApr 5, 2026, 12:27 AM
githubUrlhttps://github.com/neomjs/neo/issues/9710
authortobiu
commentsCount1
parentIssue9673
subIssues[]
subIssuesCompleted0
subIssuesTotal0
blockedBy[]
blocking[]
closedAtApr 5, 2026, 12:27 AM

[SQLite VSS Migration] 100% Offline Markdown Tensor Chunking

Closedenhancementaiarchitecture
tobiu
tobiu commented on Apr 4, 2026, 11:05 PM

Description

Phase 2 of the Memory Core overhaul. Evolve beyond containerized ChromaDB to a pure zero-dependency SQLite Vector DB driven by local qwen3-embedding.

Acceptance Criteria

  • Move Vector Storage to SQLite VSS native columns.
  • Swap standard text embedding models to qwen3-embedding.
  • Build a dedicated Semantic Markdown Chunking engine (basing chunks around ## blocks and JSON keys, rather than AST parsing).
  • Related to Epic #9673.
tobiu added the enhancement label on Apr 4, 2026, 11:05 PM
tobiu added the ai label on Apr 4, 2026, 11:05 PM
tobiu added the architecture label on Apr 4, 2026, 11:05 PM
tobiu added parent issue #9673 on Apr 4, 2026, 11:05 PM
tobiu referenced in commit f63e98c - "feat: stabilize sqlite memory core via model decoupling and querying (#9710)" on Apr 5, 2026, 12:27 AM
tobiu assigned to @tobiu on Apr 5, 2026, 12:27 AM
tobiu
tobiu Apr 5, 2026, 12:27 AM

SQLite VSS Migration complete. Resolved type-locking requirements, implemented abstraction of TextEmbeddingService (decoupling generative and embedding contexts), and verified robust pre-filtering of attributes leveraging sqlite-vec bounds.

tobiu closed this issue on Apr 5, 2026, 12:27 AM
tobiu referenced in commit c71fc11 - "chore: add throw logic for sqlite-vec native load failures (#9710)" on Apr 5, 2026, 12:30 AM