Hello! seCall is back with a big update. We've been working hard since v0.4.0 and are excited to release v0.7.0, which includes all the changes we've made.
seCall collects and searches your past conversations with various AI coding agents like Claude Code · Codex · Gemini CLI · ChatGPT locally, and organizes them into Obsidian vaults + wikis + knowledge graphs. (All local, open source)
🖥️ Improved Web UI
seCall, which was previously CLI/MCP-centric, now has a proper web UI. (We plan to connect all terminal commands to the web in the future😁)
Redesigned Session Browser - Linear style tone + responsive design for mobile. Left panel includes sorting(date/turn count/project/agent) · mini calendar(badges with the number of sessions per day + click filter) · filters.
Rich Markdown Rendering - Mermaid diagrams, KaTeX formulas, HTML previews rendered directly from session source.
Smooth Large Lists - DOM virtualization for smooth browsing even with thousands of sessions.
Live Knowledge Graph View - Real-time simulation of nodes appearing sequentially.
🔍 Enhanced Search Quality
Default Embedding Changed toqwen3-embedding→ Vector search MRR 0.903 → 0.958, hybrid 1.000.(1024-dim drop-in, no impact on existing installations)
Improved Foreign Word Phonetic Search - Filled the gap in searches with mixed Korean and English like "리프레시 ↔ refresh" → R@5 0.878 → 0.944. Expanded morphology + original token + phonetic alias using OR/prefix.
Normalized Wiki Semantic Search - Fixed a bug where query embedding path inconsistencies resulted in random ranking, ensuring accurate page placement.
Reduced Search Noise - Softly downgraded short observational sessions + filled vector result snippets.
🕸️ Knowledge Graph & Wiki
Graph Insights - Finds unexpected connections and knowledge gaps across projects.
Improved Related Session Ranking - More accurate recommendation through common neighbor weighting (Adamic-Adar).
Diversified Wiki Generation Backend - Choice of Ollama / Ollama Cloud / LM Studio / Codex + real-time progress streaming.
🧩 Stability - Full Review of Parsing Engine
Reviewed the session collection parser from start to finish and fixed 21 issues.
Blocked workflow byproducts from being incorrectly collected as sessions.
Fixed data integrity issues such as double token counting, loss of parallel tool output, silent session omission due to specific characters, and crashes caused by non-ASCII names.
Improved Codex/Gemini/ChatGPT parser date, token, and text parsing, and cleaned up ANSI color codes in tool output.
⚡ Performance
Optimized vector scanning (norm pre-calculation + SIMD + int8 cache, lossless)
Reduced large knowledge graph hangs.
🚀 Installation / Upgrade
# macOS
curl -fsSL https://raw.githubusercontent.com/hang-in/seCall/main/install.sh | sh
# Windows PowerShell
irm https://raw.githubusercontent.com/hang-in/seCall/main/install.ps1 | iex
Existing users can simply reinstall. (If using mcp, please make sure to stop secall.exe in the background before reinstalling.)
⚠️ Important Upgrade Notice for Existing Users
The default embedding has changed from bge-m3 to qwen3-embedding:0.6b since v0.6.5. To maintain search accuracy, please do one of the following:
GitHub: https://github.com/hang-in/seCall - Feedback and contributions are always welcome! Thank you for reading! 🙏
