# AI powered searching of notes?

**URL:** <https://discourse.joplinapp.org/t/ai-powered-searching-of-notes/33781>\
**Category:** Features\
**Created:** [19 November 2023 16:27 UTC](https://discourse.joplinapp.org/t/ai-powered-searching-of-notes/33781 "2023-11-19T16:27:46Z")\
**Posts on this page:** 1\
**Showing post:** 14

<div class="post-metadata">

**Author:** ![shikuz](https://yyz2.discourse-cdn.com/flex028/user_avatar/discourse.joplinapp.org/shikuz/32/21839_2.png) [@shikuz](https://discourse.joplinapp.org/u/shikuz)\
**Post date:** [23 November 2023 23:43 UTC](https://discourse.joplinapp.org/t/ai-powered-searching-of-notes/33781/14 "2023-11-23T23:43:25Z")

</div>

I recommend getting familiar with the [readme](https://github.com/alondmnt/joplin-plugin-jarvis#readme) and [guide](https://github.com/alondmnt/joplin-plugin-jarvis/blob/master/GUIDE.md) in order to understand the features that you're trying to use better.

It's worth noting that:

1. The `Chat` command is unrelated to your notes. It's just like ChatGPT, but your conversation is stored in a note, and whatever you enter in that note can be used for context.

2. `Chat with notes` adds to the context of the conversation (hopefully-)relevant sections from your notes, so that you can ask the AI about them. For example, if you have many travel logs you could ask Jarvis: "What was the thing I liked the most in our trip to Georgia?" and it will use the appropriate note to answer the question. There are additional usage examples from users in this [thread](https://discourse.joplinapp.org/t/plugin-jarvis-ai-assistant-v0-7-0-2023-08-31/28316).

3. `Semantic search` is a different thing. It uses AI, but doesn't follow instructions. See [this video](https://www.youtube.com/watch?v=fFt4kR4ntAA) for an introduction on the subject.

4. Something we didn't mention yet, but could be relevant when searching for new information, is `Research with Jarvis`, which searches for summaries of academic papers online (see also [this post](https://medium.com/@alondmnt/ai-powered-literature-review-6918ee180304)). This is not the [state of the art](https://elicit.com) for this kind of research, but still worth mentioning.

All of these are described briefly in the documentation I linked to, BTW.

It's also important to note that (1), (2) and (4) depend on a powerful LMM to get intelligent responses, GPT-4 or Claude recommended for complex prompts, while (3) can work with the default offline model reasonably well (the online models are still a bit better).

In general, Jarvis is not an agent (yet!), meaning that it doesn't perform _operations_ in the app and on your notes based on free-text, such as: "list all tags in notebook X" (a prompt that should run an API call to Joplin to get all tags and analyze them), "move all notes from X to Y" (another API call to Joplin), or "_create_ a new note about heat insulation" (another API call, you get the idea...).[\*] What it has access to is an indexed database of note sections based on embeddings. Once information is retrieved, a LMM can analyze it along with your prompt. What it can output is text that is inserted into the currently open note. It has its limitations and strengths.

[\*] One exception is auto-tagging, a specific operation that is implemented as a feature in Jarvis.

---

_[View the full topic](https://discourse.joplinapp.org/t/ai-powered-searching-of-notes/33781)._