February 7, 2026

Obsidian AI Agent with the Google Gemini Scribe Plugin For Academics

AI agents can greatly help you clean up, organise and enrich your notes. They are highly flexible and have internet access for additional information. The Gemini Scribe plugin makes Google’s Gemini models available in Obsidian. This plugin is easy enough to use for anyone, and Google offers a free tier, allowing some usage free of charge. This article explains how to configure the plugin and how academics and knowledge workers can use AI agents.

🗣️ I would like to thank Ric Raftis from the Effortless Academic Community for making me aware of this amazing plugin.

Learn the foundational systems researchers are rarely taught: from powerful note-taking and knowledge management methods to modern lit review tools, ethical AI and productive research workflows.

Step 1: Get an API Key for Google Gemini

An API key is simply a short string that uniquely identifies you. The plugin needs this key to link Gemini usage to your account. This is necessary because if the usage accrues any cost, it will be linked to you. The API key is a little bit like a credit card number and should be kept secret. However, you won’t need to pay right away, as there is a free tier available for using Google Gemini at a low rate.

  1. Go to Google Gemini and create an account.
  2. Go to Google’s AI Studio and select the API Keys tab.
  3. Click the “Create API key” button on the top right. Give the key any name and hit “create key”.

4. If you want, you can already set up billing. This will allow you to work with bigger vaults and run more queries per minute.

5. Save the API Key; we will need it in the next step. Avoid storing it in clear text on your hard drive.

Step 2: Installing and Setting Up the Gemini Scribe Plugin

Next, head to Obsidian to install the Gemini Scribe plugin from the community plugins and set a few settings. Here is a list of settings you should change:

The only necessary setting is the API key. Everything else is optional.

Step 3: Initialise the plugin context (AGENTS.md file)

Next, the Gemini agent needs to establish a context, which is just a file called Agents.md describing your vault settings, rules and conventions. It will greatly help the AI agent to do a better job and understand you more intuitively. And the best part is that it is mostly automated. Simply hit Command-P and type in “Gemini Scribe: Open Gemini Chat” to start the chat. When you first started, it might have been tucked away in the right sidebar. Click the icon, then select “Move to new window” for convenience. Here is how it looks the first time:

In order to initialise the plugin, hit the green button “Initialize vault context” to get the agent started. Depending on your vault’s size, this may take a minute or two. Now go to the plugin state folder (gemini-scribe, if you did not rename it in the settings) created in your vault, and open the AGENTS.md file. Here is how it looks for my PKM vault:

This file is editable, and you should edit it to match your vault. For instance, I added a few more descriptions about my notes being atomic (read on the benefit of atomic notes), and a type property which defines whether a source note refers to a book, a website, or something else. The majority of the above file is, however, AI-generated. Whenever you feel the AI isn’t quite getting the structure of your vault, go back to the Agents.md and modify it.

At this point, your agent is fully set up and ready to go. Before we start, it is useful to understand what the agent is capable of doing.

What can Gemini’s Scribe AI agent do in Obsidian?

The documentation provides a thorough list of the agent’s capabilities. These are:

  • Reading, writing and deleting files
  • Running Google queries
  • Fetching data from any URL and analysing the output
  • Searching for files and their content

More importantly, as an AI agent, it can combine these capabilities into a sequence of actions to accomplish complex tasks and help with grunt work or synthesis (i.e., pulling related information from multiple sources). Let’s look at a few use-cases for academics and knowledge workers next.

Use Case 1: Organising & Enriching

Don’t think of AI as just a tool to answer a question. Think of it as an assistant that can help you with low-level multi-step tasks. Here is an example in which AI must aggregate widely available information into a new note. Every so often, I get my blood results checked and receive a dozen or so values (e.g. B12 vitamin, iron, cholesterol), and I want to (1) understand what these markers indicate, (2) what the consequences are of a marker being too high or too low, and (3) learn how they influence each other. This is not something that can be done with a single prompt, but we can use an AI agent to help us build that understanding much faster.

Here is my first prompt:

Create a new note "Blood Markers" and populate it with all blood markers you can find in @"Blood Results comparison 2022 - 2023". For each marker write down in very brief sentences what it is responsible for (e.g. for nerve cells) and what too much and rtoo litle of it entails. Add my value from 18.Dec 2023 into this note.

Note that by adding an @ character you can refer to a specific note in your vault. The UI makes it very easy to select a specific note. And here is the result:

While the agent was working, you could see it was searching for high and low values of various blood markers to generate the table in the results. The table had about 40 entries, and you can only see a few in the screenshot. This saved me at least an hour of time trying to search for what these different markers mean! Note that this entire note was created by the agent, and I didn’t have to copy and paste anything from the chat to my notes.

You can imagine the agent perfectly capable of creating a separate note for each marker and diving deeply into what the values mean and how they correlate. Essentially, creating the content that is relevant to me from all of the content found on the internet. The AI agent here acts as a “filter of information”. It filters the internet to the content relevant to my brain at the specified level of detail. This is what makes it so powerful and flexible

Use Case 2: Filtering a Journal Feed

In this use case I ask the AI to go through the newest issue of Nature and filter out relevant articles based on the contents of my notes. Most journals have a so-called “RSS feed”, which is a machine-readable version of the articles they publish and is perfectly suited for AI analysis. Here is my prompt:

Look at this URL https://www.nature.com/nature.rss Find all articles that you think might be relevant to me based on my notes. Your output should contain a link (Use the DOI identified to construct the link) to the article linked at this URL and mention some of my notes in the description as markdown links to signify why you think this article is relevant.

And here is the result:

Based on different books and notes, the agent found articles that I might find relevant. While this is arguably not the most common use case, it highlights an exciting feature of AI agents: access to any type of external data or web service that can be combined with your notes and filtered to your needs. For instance, this workflow would not be possible with ChatGPT, NotebookLM or any other advanced AI, because they don’t have access to my notes and uploading them every time is simply not practical. When deciding how to use the agent, always consider its leverage over “regular chatbots”, which is your notes and the context they provide.

Use Case 3: Organising your Notes

The agent is also well-suited to organising your content, as it can read, write, and delete files. (Be careful about giving it too much freedom, as a mistake can lead to content loss. If you lose something, hit Ctrl/Cmd + P and type: “File Recovery: Open Local History”. In this workflow, I have the agent analyse my map of content on various plant functional traits. So far, this list was quite dirty, but with a simple prompt, the agent read all the ~30 notes in the list and created a wonderful overview table.

More importantly, it reads each note before creating the file and summarises its content. In some cases, it even linked a few additional notes into the description. This use case exemplifies that a lot of the housekeeping in a personal knowledge management vault will soon be done by AI.

Obsidian Gemini AI sometimes gets stuck or times out

Having experimented with the tool on many occasions and tried to write files in bulk, I ran into bugs. Sometimes the AI tries to write a file and asks for permission. Sometimes it succeeds, and other times it fails to overwrite existing files or edit them and times out instead. I think this is to be expected with a fairly new open-source project like the Gemini Scribe. However, having also worked with Obsidian’s Co-pilot, I find the agent still more powerful and intuitive than Copilot.

Sometimes restarting Obsidian helps, sometimes the AI simply does not have permissions to write the file, and it seems to be a file system issue. My feeling is that this will eventually be resolved. You can also try OpenAI’s Codex and Anthropic Claude Code, which run directly on the file system and don’t encounter these issues, though they are a bit more technical than Gemini Scribe.

Use Case 3: Rewriting and editing text

One of the simplest use cases involves selecting text and prompting the AI to rewrite it. The difference in using ChatGPT or another stand-alone LLM is that Gemini is aware of the context and the links contained in your text and can therefore do a much better job at rewriting it. Here’s an example:

While this is probably not a use case that you would want to follow to publish your papers (see ethics discussion), it can be useful in creating internal documents and summaries of your own work, and thus improving the connectivity of your notes, which in turn helps with creative ideas. Another way to help improve the connectivity is by using Obsidian bases.

Obsidian AI commands for repeatable workflows

Whenever you encounter a use case that will be repeated, it’s worth encapsulating it as a command. Then all you need to do is tell Gemini to run this specific command and point it to the content to use as input. For instance, you might have a command that automatically creates a map of the content of all concepts related to the note that you’re editing, or you might want to convert a list into a markdown table, or generate a piece of coherent academic writing from your bullet point notes.

To add new commands, open the AGENTS.md file in the gemini-scribe folder, create a new header titled “Commands” at the bottom, and use subheaders to add prompts that will become the commands. Here is an example for the “fill out and relate” command.

If you are unsure how to describe what you’re doing, it’s usually a good idea to break down your prompt into a few parts:

  1. Role: This helps the AI to align with your expectations
  2. Tone: This tells it exactly how to create the output and controls how verbose it is.
  3. Action: This breaks down the task into a series of steps the agent will do
  4. Rules: Here you can add some do’s and don’ts for the AI to further constrain its behavior, if not included in “Actions”.
  5. Format: if you have very specific output requirements, you can use a separate point for that. Otherwise, just include this in “Actions”.

Just following this outline, you will create fairly effective instructions. But of course, experimenting is key to fine-tuning the command to your needs. The process is identical to the one you would use to specify a custom GPT bot. Check Step 5 in the CustomGPT tutorial for more information.

Example Command: Find and Relate Command

The goal of this command is to find any relevant information in my vault whenever I come across a new topic. Since I have been working on my subject for many years, my vault certainly contains papers and notes I took a long time ago and have since forgotten. This command will help me set up my notes and find all relevant connections and papers.

Here is the prompt that I have used to find and relate ideas to a new concept:

### Fill-out-and-Relate
**Role:** Academic librarian
**Tone:** Factual, concise, academic language. Use jargon where needed. 
**Action:** Run these three steps  
1. Take the title of this note and any information provided in the chat to identify the topic of this note. 
2. Use either a search or your knowledge to define this topic and place it in a callout at the top. Mark the callout with "AI-definition"
3. Now scan the primary sources folder and the collections folder to find any information that should be relevant to this topic. Create two lists one titled "Sources" and one "Conecepts". Put a list of links and give one-sentence explanation of how that other note is related to this note and why it is important. 
4. If not pointed to an existing note, create a new note with the output using the template file: `__Templates/Stub.md

Just copy and paste this prompt under the commands section at the bottom of the Agents.md file. You possibly need to modify that prompt a bit because under (3) I mention the primary sources and collections folder that you may not have, and under (4) I point to a template that you also likely won’t have. (The note-taking course goes into detail about how I organise my vault).

Then ask Gemini to “Run the fill out and relate command on the topic of <your topic>”. Here is the output for a topic I am studying: “Drivers of functional diversity”.

When I first saw this note, it completely blew my mind. Despite this topic being new, I already had 6 papers on it and quite a few notes on related topics. This approach makes it much easier to learn and understand a new topic because it is now connected to so many other things in your vault. And the bigger your vault grows, the more these existing connections compound to better ideas that AI can help you discover.

A word of warning: Creating content with AI does not equate understanding or learning. It seems like everything is connected, and the information is there. However, it is only in your digital mind, not your physical mind. Text on a hard drive is “dead”, it does not produce new ideas. Use the AI output to explore and double-check(!) these connections yourself and write in your own words what they mean to you. Only this process will ensure that you learn and anchor this new concept within your domain of knowledge and can subsequently use it to generate new ideas. Time spent understanding is what gives “life” to information.

Using Gemini Scribe Plugin With Hover Editor

The Hover Editor plugin is one of my favourite plugins (see the full list of Obsidian plugins) because it lets you pop any content or plugin into a hovering window that stays on top. Especially when working with an AI chat, this is very handy. Here is how it looks:

To move the Agent window to a new Hover editor, click the three dots on the top right of the Agent window and select “Convert to Hover Editor.” Keep in mind that you are not limited to a single window but can have as many as you like. There is an option called “Split right/down” to create additional windows in the three-dot menu in the top-right.

Alternatives to Gemini Scribe in Obsidian

If you don’t want to use Google’s Gemini, you can switch to the Smart Composer plugin, which achieves a similar result but connects to other AI providers as well. Since everything is markdown-based, you can also use OpenAI Codex or Anthropic Claude Code. There is also nothing stopping you from using all of these tools in parallel.

Free AI Skills for Academics Guide

Is Obsidian AI with Gemini free?

To a small extent, yes. However, in my large vault, I quickly ran up against the limits on how many requests per minute you can make, which are easily exhausted. After upgrading to a paid tier and working for many hours on this article, I ended up spending about 0.30$.

You can view your current usage and costs in the AI studio usage and billing tab.

Obsidian Copilot vs Gemini Scribe

Copilot is more of a chat agent that can help you answer questions regarding your vault, while Gemini’s scribe is an AI agent that is built to run certain tasks. That makes it more powerful than Copilot. However, Copilot is a for-profit project and contains fewer bugs. Most importantly, PDF and attachment indexing is included with Copilot, which I wasn’t able to get to work with Gemini Scribe (It is considered “experimental” by the makers at the moment and also involves uploading your files to the cloud). However, I found Gemini Scribe somewhat more intuitive in the way it works, for these reasons:

  1. The Agents.md file allows you to share expectations with the AI.
  2. While the agent operates, you can see exactly which actions are taken.
  3. CoPilot often gets overloaded by too many files and throws an error regarding context length. This never happened to me with the Gemini Scribe plugin.
  4. I always found the option of chat vs vault QA in CoPilot confusing and Gemini Scribe appears to always be aware of my entire vault.

Gemini Scribe vs SmartConnections

Smart Connections is a plugin that helps you find notes that are semantically similar to the currently open note or a query. This allows you to search your vault and discover new connections, as the name suggests, and it can run offline. It is a lightweight, free alternative if your primary goal is to search Obsidian notes. However, if you want to generate content, you will need Gemini Scribe, Copilot or any other Obsidian AI plug-in. Here is a review on how to use these plugins:

Summary: Gemini Scribe AI and Obsidian

With Gemini’s Scribe plugin, Obsidian’s AI got another significant upgrade. In my opinion, it is one of the first times AI has crossed the threshold from a tech demo to something truly powerful and useful for personal knowledge management. Moreover, it allows me to do things I would not have done before, either because I was not aware of certain notes and could not find them, or because I was simply too lazy.

Most importantly, this development indicates where AI in personal knowledge management is headed. AI will allow us to do less and less organising and will especially benefit people who don’t have a clear system or many templates set up to automate their PKM. Moreover, AI will jumpstart your memory by providing old forgotten connections and linking parts of your vault that you would not necessarily think of.

Free 21-day course: Obsidian for Academics
Organise your research digitally, find every PDF you read and every claim you highlight and stay on top of large research projects. 21 daily lessons and helpful links. Free.

Effortless Academic helps students and professionals build better workflows for note-taking, writing, and AI-assisted research.

Join the Newsletter & Get a Free 21-day Course

A free 21-day course to teach you how to remember every paper you read, generate new ideas faster and leverage the best workflows & AI tools to maximise your productivity. Recieve one daily email with a clear lesson and homework to get started on information management with Obsidian.

Start your free 21-day course

This course is for any academic, student, researcher or knowledge worker who wants to learn topics faster, keep track of sources and write more efficiently. Every day, you’ll get one email with a dedicated lesson.