Disclaimer
Is this project open source? Yes
Is this project completely free? Yes
Is this project vibe-coded beyond the author’s ability to comprehend how it works? No
Community Directory:https://community.obsidian.md/plugins/lantern
Hi,
A new, simple search/AI plugin using self-hosted qmd (Markdown semantic search/indexer) and OpenAI API compatible server (llama.cpp, LM studio …).
The idea: not to get lost in the vault - hence the name. 
The primary focus was simplicity and privacy, but if anyone wants to use hosted models - the plugin should support that too.
It offers lexical/semantic search and a simple agent loop with very few tools, and prompts designed for vault only (or enabled references) information retrieval with citations.
What the agent does:
- Search semantically and lexically using qmd (good embedding model is key here! Check README) to identify relevant notes in the vault and/or enabled external references.
- Read relevant parts of the found sources to synthesise reply.
- Create a reply, using citation marks for every stated fact.
- Create a citations footer for every citation mark used.
The qwen3.6-27b, 35b-a3b and gemma4-26b-a4b (tested at Q8) do surprisingly good job.
The plugin optionally supports external (relative to the vault) references: other qmd indexed collections or web search: Perplexity (API key required) and Exa (API key optional) supported.
I designed, extended and polished it for my needs and taste, but seeing how good it can be with finding information and sources - I’ve decided to publish it for others.
It’s mostly for power users who can handle qmd and local LLM server and who have powerful enough hardware. I’m using it on a MacBook Pro M1 Max 64 GB, with llama.cpp and qwen3.6-35b or gemma4-26b models - and the results are great. Check the description/github for more details.
Enjoy!