Release 1.4.0 · green-dalii/obsidian-llm-wiki

What's New in 1.4.0

This release consolidates all Phase 1 robustness work and passes Obsidian review bot validation.

Schema Layer — schema/config.md, the third layer from Karpathy's design. Human-editable config that governs LLM operation on the Wiki (naming
conventions, page templates, classification rules)
Auto-Maintenance — file watcher + periodic lint + startup health check. All default OFF to avoid surprise API costs.
JSON Output Mode — forces response_format: { type: "json_object" } for all structured LLM calls, eliminating malformed JSON at the source
Ingestion Report Modal — structured summary window after each ingest: created/updated pages, failed items, contradictions found
Hierarchical Index — LLM-generated tree-structured index grouped by type, with automatic flat fallback for large wikis (>25 pages)
Schema Suggestion — LLM analyzes Wiki health and proposes schema improvements

Network Resilience: 5-min timeout + exponential backoff retry (max 3) for transient connection errors
Fault Tolerance: per-entity/concept try-catch with auto-retry; single page failure no longer aborts entire ingestion
JSON Parsing Robustness: state-machine content-quote escaping + LLM repair pipeline handles unescaped quotes, trailing commas, missing commas
Index Generation Hang: large wikis bypass LLM and use fast flat index; max_tokens 2000→4000
API Throttling: 300ms delay between LLM calls to prevent connection errors
Frontmatter Wikilinks: source file references now use [[path]] format for clickable Obsidian links
Progress Toast: auto-dismisses when ingestion completes

Full Changelog: CHANGELOG.md