OmniOwn v0.1.4

OmniOwn v0.1.4 Release Notes

Release Date: 2026-06-25

Overview

v0.1.4 introduces a two-stage AI search pipeline with intelligent query analysis and strategy selection. Enhanced search accuracy through LLM-powered query understanding, combined with structured validation and document library context awareness.

Key Features

🤖 Two-Stage AI Search Pipeline

• Stage 1 (Query Analysis): LLM-powered preprocessing that rewrites queries, extracts keywords, and detects intent, category preferences, filetype filters, and time-range constraints
• Stage 2 (Strategy Selection): Intelligent strategy selection with JSON Schema validation via Zod, ensuring only valid strategies are executed

🔍 Enhanced Search Context

• Document Stats Cache: Total document count and category distribution injected into the v2 prompt system
• 60-second TTL Cache: Automatically invalidated on import and watch events
• Tiered Result Merging: FTS results prioritized; non-FTS results capped at 5 when FTS exists to reduce noise

🌐 Improved Internationalization

• All system prompts converted to English instructions with mixed CN/EN few-shot examples
• Better JSON compliance for LLM outputs
• Modularized prompt architecture for easier maintenance and reuse

Changes

Added

• Query Analysis Prompt Module (server/src/prompts/query-analysis.prompt.ts)

  • Two-phase LLM query understanding: rewrite + intent/category/filetype/time-range detection
  • Structured JSON output for programmatic processing

• Zod JSON Schema Validation

  • Strategy selection output validated against StrategyResponseSchema
  • Enum-checked strategy names, minimum 1 strategy requirement
  • Descriptive error messages on validation failure

• Document Library Context

  • getDocumentStats() service for gathering library metadata
  • Automatic cache invalidation on document changes
  • Context injection into v2 prompt as [Document Library Info] block

Fixed

• v2 Context Never Injected: getDocumentStats() now properly called in selectStrategies when variant is 'v2'
• Search context availability verified and properly passed through the pipeline

Improved

• Prompt Modularization: Prompts extracted from ai.service.ts into dedicated prompts/ module

  • search-strategy.prompt.ts with v1/v2 variants
  • 6 few-shot examples per strategy
  • Context injection and intelligent fallback
  • index.ts barrel exports for clean imports

• Result Quality: Tiered result merging reduces noise while preserving browsing capabilities

Technical Improvements

• ✅ Replaced bare as StrategyCall[] type assertions with proper Zod validation
• ✅ Cache invalidation integrated with import and watch workflows
• ✅ Prompt system isolation enables easier testing and maintenance
• ✅ Better error messages for debugging search issues

Breaking Changes

None — This release maintains backward compatibility with v0.1.3.

Migration Guide

No action required. Simply upgrade to v0.1.4 and enjoy improved search accuracy.

Performance

• Query analysis adds ~200-500ms depending on LLM endpoint
• Document stats cache reduces overhead on repeated queries
• Result filtering improves UI responsiveness with large document sets

Related Documentation

• Architecture Overview — System design and component interactions
• Search Strategy Guide — CLI usage and search capabilities
• Development Guide — Setup and contribution workflow

Known Limitations

• LLM query analysis depends on API availability
• Category detection accuracy varies by document naming conventions
• Time-range extraction requires ISO 8601 or common date formats

OmniOwn v0.1.3

Release v0.1.3

OmniOwn v0.1.2

🎉 OmniOwn v0.1.2 — Windows Installer Refresh

This release keeps the v0.1.1 Windows watcher/config fixes and adds both Windows installer formats, matching the Windows assets style from v0.1.0.

📦 Windows Downloads

• OmniOwn_0.1.2_x64-setup.exe — NSIS setup installer for quick Windows installation.
• OmniOwn_0.1.2_x64_en-US.msi — MSI installer for Windows deployment workflows.

🐛 Fixes Included

• Fix packaged startup config generation so default Windows library paths are TOML-safe.
• Restart the file watcher after saving path settings so custom library directories take effect.
• Ensure packaged omniown watch receives the configured library path instead of falling back to .\library.

🔄 Changes Since v0.1.1

• Release packaging now builds both Windows MSI and NSIS setup.exe installers.
• This release intentionally publishes only Windows installer assets.

OmniOwn v0.1.1

OmniOwn v0.1.1

OmniOwn v0.1.0

---

🎉 Omniown v0.1.0 — First Release

This is the initial release of Omniown, a local-first personal knowledge base with AI-powered search, built as a three-tier full-stack application.

🏗️ Architecture

Rust CLI (text extraction + file pipeline + MCP) → Node.js API (Express + Prisma) → Vue 3 frontend + Tauri v2 desktop shell

✨ Highlights

• FTS5 Full-text Search — SQLite FTS5 virtual tables with automatic trigger-based sync, delivering millisecond-level queries.
• AI Multi-Strategy Search — LLM-driven intent analysis selects the optimal combination from 8 search strategies, executed in parallel with deduplication.
• MCP Server — Exposes 4 tools (search_documents, get_document, list_documents, get_status), allowing AI clients to directly access your local knowledge base.
• Multi-Format Text Extraction — Supports plain text, Markdown, HTML, code, JSON/YAML/TOML/CSV, PDF, DOCX, and XLSX.
• File Import Pipeline — SHA256 deduplication, auto-classification (public/private), and interactive same-name conflict resolution.
• Tauri v2 Desktop App — System tray + floating panel, auto-spawns a Node.js child process as the API backend.

📦 What's Included

• CLI text extraction & import pipeline
• Node.js REST API (Express 5 + Prisma 5, TypeScript strict)
• Vue 3 frontend (Pinia stores, Vite 6)
• Tauri v2 desktop shell (system tray + floating panel)
• MCP server for AI client integration
• TOML-based configuration (inbox/library paths, LLM API settings)

🔄 Changes Since Pre-release

• Architecture refactored from a Rust monolith to a three-tier full-stack design
• Database module streamlined — migration, classifier, and storage logic inlined
• Frontend refactored from direct API calls to Pinia stores + service layer
• API response fields converted from snake_case to camelCase (matching Prisma native output)

❌ Removed

• Embedding-based vector search (replaced by AI multi-strategy search)
• File-watching sentinel mode (replaced by Node.js API + manual CLI import)

🐛 Fixes

• Frontend/backend API response format mismatch (wrapper + field naming)
• CI package-lock.json gitignore causing npm ci to fail
• Release CI sidecar path mismatch due to missing --target flag on tauri-action
• macOS x86_64 build switched to macos-15-intel runner
• GitHub Actions upgraded (Node.js 20 deprecation): checkout@v6, upload-artifact@v7, download-artifact@v8, setup-node@v6
• Release CI: write permissions, artifactPath, and packaging type configuration fixes

---