1.6.0

Added

• Export similar notes for active note (#47): New command Similar Notes: Export similar notes for active note writes the current note's results to .obsidian/plugins/similar-notes/similar-notes-export.json, letting external tools (e.g. coding agents) reuse the plugin's similarity search. Each result carries path, title, score, excerpt, and linked.

Changed

• Already-linked notes now appear in Similar notes: notes the active note already links to are no longer hidden — they show ranked by score alongside everything else, marked with a small muted "linked" tag.
• Built-in models on mobile now warn about memory use: on lower-memory phones, built-in on-device models can crash Obsidian, so the Built-in model settings now recommend a remote provider (Ollama, OpenAI, or Gemini). High-memory phones and tablets can still use built-in models.

Improved

• Bug reports now include CPU and memory: the "Copy environment info" output adds CPU core count, architecture, and total RAM (desktop only) to help diagnose memory- and performance-sensitive issues.

Fixed

• Built-in model no longer crashes when indexing large notes (with GPU acceleration off): a large note's chunks are now embedded in smaller batches instead of one pass that could exceed the on-device engine's fixed memory limit (previously surfaced as a bare number like 8934496 in Errored files). Retry affected notes after updating.
• Better recommendations for long, multi-topic notes on large-context models: chunk size is now a focused semantic size regardless of the model's context window, so a single topic's signal is no longer averaged away in coarse chunks. Run a full reindex to apply the finer chunking.
• Ribbon icon hidden state now sticks across restarts (#50): a hidden Similar Notes icon no longer reappears on every launch, and clicking the icon during startup no longer opens an empty pane.
• "Copy environment info" now reports the correct provider and settings: the report reads your settings at click time (no more stale model after switching), and shows the actual provider/model for OpenAI and Gemini instead of always Built-in (...), with a Server URL line to distinguish OpenAI, OpenRouter, and local servers.