Update Research Notes with AI in Obsidian

Use an AI agent in Obsidian to refresh tech research notes with live web data.

Introduction

30 Second Summary

You have dozens of tech research notes scattered across your vault. Over time, frameworks release new versions, best practices shift, and your notes quietly go stale. Manually checking and rewriting each one is tedious enough that most people just never do it.

In this project, you will set up an AI agent inside Obsidian that can search the web for current information and update your tech research notes on demand.

What You'll Build

A complete AI-powered research update workflow using the Gemini Scribe plugin, which gives your Obsidian vault a built-in AI agent with web search capabilities.

By the end of this project, you'll have:

Gemini Scribe installed and configured with a free API key, ready to update notes on demand.
A tech research note successfully updated with current web information using the AI agent's search capabilities.
A reusable "research-updater" agent skill that standardizes how the AI refreshes your notes every time you trigger it.
Secret Mission: Create a second skill that generates a changelog note summarizing all updates made in a session.

Are there any prerequisites?

You need Obsidian installed with community plugins enabled, and a Google account for generating your free API key. No coding experience is required.

Install Gemini Scribe and Get Your API Key

To update your research notes with current information, you need an AI agent that can read your notes, search the web, and write edits back into your vault.

In this step, you will install Gemini Scribe, an Obsidian plugin that brings a Gemini-powered AI agent directly into your vault. You will also create a free API key that lets the plugin talk to Google's Gemini models.

In this step, get ready to:

Install the Gemini Scribe plugin from the community plugins browser.
Create a free Gemini API key from Google AI Studio.
Connect the API key to the plugin so it can communicate with Gemini.

Install the Gemini Scribe plugin

Gemini Scribe is a community plugin, which means it is not bundled with Obsidian by default. You will install it from the community plugins browser inside Obsidian's settings.

In Obsidian, click the gear icon in the bottom-left corner to open Settings.
In the left sidebar, click Community plugins.
Click Browse.
In the search bar at the top, type Gemini Scribe.
Click on the Gemini Scribe result.
Click Install.
After installation completes, click Enable.

You should see a confirmation that Gemini Scribe is now enabled in your community plugins list.

What is Gemini Scribe?

Gemini Scribe is an AI agent plugin that connects your Obsidian vault to Google's Gemini models. Unlike basic AI plugins that only chat, Gemini Scribe can read your files, edit your notes, search the web via Google Search, and follow multi-step workflows you define.

The built-in web search is what makes it perfect for keeping research notes current. Other Obsidian AI plugins can only work with what the model already knows, which may be months out of date.

Don't see Gemini Scribe in the search results?

Make sure community plugins are turned on. Go to Settings, then Community plugins, and confirm the toggle at the top is enabled.

If you still cannot find it, check that your Obsidian app is up to date by going to Settings, then General, and clicking Check for updates.

Help me troubleshoot why Gemini Scribe isn't appearing in my Obsidian community plugins browser.

Create your free Gemini API key

Gemini Scribe needs an API key to authenticate with Google's Gemini models. You will create one for free through Google AI Studio.

Open your web browser and go to https://aistudio.google.com/apikey.
Sign in with your Google account if prompted.
Click Create API key.
If prompted to select a project, accept the default project that Google creates for you.
Copy the generated API key to your clipboard.

You should see a long string of characters displayed on screen. This is your API key.

Is this really free?

Yes. The Gemini API free tier includes free input and output tokens, plus 500 Google Search requests per day. For manually updating research notes on demand, you will never come close to these limits.

Having trouble creating an API key?

Make sure you are signed into a personal Google account. Some workplace or school Google accounts have restrictions that prevent API key creation.

If the page asks you to accept Terms of Service, accept them and try again.

Help me troubleshoot creating a Gemini API key in Google AI Studio.

Connect your API key to Gemini Scribe

Now you will paste your API key into Gemini Scribe's settings so the plugin can authenticate with Google's servers.

Back in Obsidian, open Settings (gear icon in the bottom-left corner).
In the left sidebar, scroll down to Community plugins and click Gemini Scribe.
Find the API Key field.
Paste your API key into the field.
Confirm the Chat Model field is set to gemini-flash-latest. This is the default and does not need to be changed.
Close the Settings window.

Why gemini-flash-latest?

The gemini-flash-latest setting automatically routes your requests to Google's newest Flash model. This means you always get the latest improvements without manually updating the model name.

Is my API key stored securely?

Gemini Scribe stores your API key using Obsidian's SecretStorage, which keeps it encrypted and separate from your vault files. All communication with the Gemini API happens directly from your machine to Google's servers. No third-party servers are involved.

Your AI agent is now connected and ready to work. Next up, you will open the agent panel and run your first research update on a real note.

Explore the Agent and Run Your First Update

Your Gemini Scribe plugin is installed and connected to the Gemini API. Now it's time to meet the AI agent that lives inside it.

In this step, you'll open the agent's chat panel, ask it to update one of your tech research notes with current information from the web, and approve the changes it proposes.

In this step, get ready to:

Open the Gemini Scribe agent chat panel.
Prompt the agent to update a tech research note with web-sourced information.
Review and approve the proposed changes.

Open the agent panel

Gemini Scribe adds a dedicated chat panel to Obsidian where you interact with the AI agent. Think of it like a messaging window where you give the agent instructions and it reports back what it finds or wants to change.

Look at the left ribbon in Obsidian (the narrow vertical toolbar on the far left).
Click the brain-circuit icon to open the agent chat panel.

Use the command palette instead

If you can't spot the icon, press Cmd+P (macOS) or Ctrl+P (Windows) to open the command palette. Type Gemini Scribe: Open Gemini Chat and press Enter.

A chat panel appears on the right side of your Obsidian window. This is where you'll send all your instructions to the agent.

Don't see the brain-circuit icon?

Make sure Gemini Scribe is enabled. Go to Settings, then Community Plugins, and check that the toggle next to Gemini Scribe is on.

help me find the Gemini Scribe agent panel in Obsidian

Prompt the agent to update your note

The agent can read your notes, search the web using Google Search, and propose edits. You'll combine all three capabilities in a single prompt.

In the main editor area (not the chat panel), open one of your tech research notes that you know is outdated or could use fresh information.
Click into the chat input at the bottom of the agent panel.
Type or paste the following prompt, then press Enter to send it:

Read my current note and search the web for the latest developments on this topic. Update the note with any new information, keeping my existing structure and adding citations for new facts. Flag anything that has changed significantly.

What does this prompt do?

This prompt triggers three agent capabilities in sequence:

"Read my current note" tells the agent to use its Read Files tool on whatever note you have open in the editor.
"Search the web" triggers the agent's Web Search tool, which uses Google Search through the Gemini API.
"Keeping my existing structure" prevents the agent from reorganizing your note. It will slot new information into your existing headings.
"Adding citations" ensures new facts come with source URLs so you can verify them later.

The agent will take a moment to read your note, search the web, and compose its proposed changes. You'll see its progress in the chat panel as it works.

What is Web Search vs Deep Research?

The agent has two research tools. Web Search does a quick Google search and is usually sufficient for updating individual notes.

Deep Research performs a comprehensive multi-source investigation. It's better for creating entirely new reports from scratch. For this project, the agent will use Web Search.

Review and approve the changes

The agent never edits your notes silently. When it wants to make changes, it shows you exactly what will change and waits for your approval.

Look at the diff view that appears in the chat panel. This shows a side-by-side comparison of your original text and the proposed edits.
Read through the proposed changes. The agent should have added new information under your existing headings, along with source URLs as citations.
If you're happy with the changes, click Approve to apply them to your note.
If you want to adjust something, edit the proposed text inline before approving. You can also type a follow-up message telling the agent what to change.

What should I see?

After approving, your note in the editor updates with the new content. You should see fresh information that wasn't there before, formatted under your existing section headings, with source URLs added as footnotes or inline links.

Agent not finding your note?

Make sure the note you want to update is open and active in the editor (not just visible in the sidebar). The agent reads whichever note is currently focused.

If the agent returns generic information without reading your note, try being more explicit: "Read the note I have open right now, then search the web for updates on the topics it covers."

help me get the Gemini Scribe agent to read and update my open note

You just updated a research note with current web information without leaving Obsidian. Next up, you'll configure the agent's permission settings so reads happen automatically while writes always require your approval.

Configure Tool Permissions for Your Workflow

In the previous step, you saw the agent read your note, search the web, and propose edits all in one flow. That works, but right now the agent can do all of this without asking for your permission first.

For a research workflow, you want the AI to gather information freely but always pause before changing your notes. Gemini Scribe has permission presets that control exactly this. In this step, you will lock down the agent so reads happen automatically and writes always require your approval.

In this step, get ready to:

Set the Cautious permission preset so reads are auto-approved and writes need confirmation.
Test the approval flow by prompting the agent to update a note and observing the confirmation modal.

Set permission mode to Cautious

Gemini Scribe offers four permission presets that control how much freedom the agent has. The one you want is Cautious. It auto-approves read operations like searching your vault and searching the web, but requires your explicit confirmation before any write touches your notes.

Open Settings in Obsidian.
In the left sidebar, scroll down to Community Plugins and click Gemini Scribe.
Find the tool permissions section.
Select the Cautious preset.
Close the settings panel.

What are the other permission presets?

Gemini Scribe has four presets in total. Read Only blocks all writes entirely. Cautious (the one you just selected) auto-approves reads but asks before writes.

Edit Mode auto-approves most writes without asking. YOLO approves everything without asking. For a research workflow where accuracy matters, Cautious gives you the best balance of speed and control.

Test the Cautious permission flow

Now you will see Cautious mode in action. The agent should search and read without interrupting you, then pause with a confirmation modal the moment it wants to edit your note.

Open one of your tech research notes that still has outdated sections.
Click the brain-circuit icon in the left ribbon to open the agent chat panel (or press Ctrl+P and search for Gemini Scribe: Open Gemini Chat).
In the agent chat, type a prompt asking it to update a specific section of your note with the latest information from the web.

Watch what happens next. The agent reads your note and searches the web without stopping to ask you. These are read operations, so Cautious mode lets them through automatically.

When the agent finishes researching, it will show a confirmation modal with a side-by-side diff of the proposed changes.
Review the diff to see exactly what the agent wants to add or change.
Click approve to accept the edits, or edit them inline before accepting.

Why does this matter for your workflow?

With Cautious mode, you get the speed of automated searching without the risk of unwanted edits. The agent does the heavy lifting of finding and synthesizing information, but you remain the final gatekeeper for what actually goes into your notes.

Not seeing the confirmation modal?

Make sure you selected Cautious and not Edit Mode or YOLO. Both of those auto-approve writes without asking. Go back to Settings and double-check the permission preset.

help me fix my Gemini Scribe tool permissions so I see a confirmation before edits

Your permissions are locked in. The agent can gather information at full speed, but every edit passes through you first. Next up, you will create a reusable skill that encodes your exact update process so you never have to re-type the same prompt.

Create a Reusable Research Updater Skill

Your tool permissions are dialed in. The agent reads freely and pauses for your approval before writing. Now you have a bigger problem to solve: every time you want to update a note, you have to write out a detailed prompt from scratch.

Agent skills fix this. A skill is a reusable set of instructions stored as a markdown file in your vault. Once you create a skill, you can trigger it with a single sentence instead of re-explaining your entire process every time.

In this step, get ready to:

Ask the agent to create a research-updater skill.
Refine the skill with specific step-by-step instructions.
Test the skill on a different research note.

Ask the agent to create the skill

Gemini Scribe stores skills as markdown files inside your vault at gemini-scribe/Skills/[skill-name]/SKILL.md. You do not need to create the file manually. The agent can create it for you when you ask.

Open the Gemini Scribe agent chat panel by clicking the brain-circuit icon in the left ribbon.
Type the following prompt into the agent chat:

Create a skill called research-updater that helps me update tech research notes with the latest information from the web.

The agent will show a diff view of the new SKILL.md file it wants to create. Click approve to accept.

What just happened?

The agent used its create_skill tool to generate a folder and SKILL.md file inside gemini-scribe/Skills/research-updater/ in your vault. The file contains YAML frontmatter (name, description) and a markdown body with instructions.

Because your permissions are set to Cautious, the agent asked for your approval before writing the file. This is the same approval flow you saw when it edited your notes.

You should see the agent confirm that the skill was created. The skill now exists in your vault as a file you can version control and edit anytime.

Agent didn't create the skill?

Make sure the skill name uses only lowercase letters, numbers, and hyphens. Names cannot start or end with a hyphen, and consecutive hyphens are not allowed.

If the agent asks a clarifying question instead of creating the skill, reply with confirmation and it will proceed.

Help me troubleshoot why the agent didn't create my research-updater skill in Gemini Scribe.

Refine the skill instructions

The skill the agent created has basic instructions. Now you will replace them with a detailed step-by-step process that tells the agent exactly how to update your notes every time.

In the same agent chat, type the following prompt to update the skill with your specific process:

Update the research-updater skill with these steps: 1) Read the current note and identify the main topics covered. 2) Search the web for each topic to find developments from the last 3 months. 3) Compare findings against what is already in the note. 4) Add new information in the appropriate sections, keeping my existing structure. 5) Add a 'Last Updated' date in the frontmatter. 6) Add source URLs as footnotes for any new claims.

The agent will show a diff view of the changes it wants to make to the SKILL.md file. Review the updated instructions.
Click approve to save the updated skill.

What does this skill encode?

Your six-step process tells the agent to read first, research second, compare third, and write last. This order prevents the agent from jumping straight to edits without understanding what has changed.

The frontmatter date and footnote citations ensure your notes stay traceable. You can always see when information was last refreshed and where it came from.

The agent should confirm the skill was updated. You now have a complete, reusable process stored in your vault.

Edit skills anytime

Skills are just markdown files. You can edit SKILL.md directly in Obsidian, or ask the agent to update it again with new instructions. The file persists across Obsidian restarts.

Skill instructions look different from what you expected?

The agent may rephrase your instructions slightly. As long as all six steps are represented, the skill will work correctly. If a step is missing, tell the agent which step to add and it will update the skill again.

Help me fix my research-updater skill instructions in Gemini Scribe.

Test the skill on a different note

The real test is whether the agent follows your defined steps instead of improvising. You will now trigger the skill on a note you have not updated yet.

Open a different tech research note in your vault that has not been updated yet.
In the agent chat, type the following prompt:

Use the research-updater skill to update this note.

Watch the agent work through your six steps in order. It will read the note, search the web, compare findings, and then propose edits.
When the agent shows the diff view of proposed changes, review the edits. Confirm that new information is placed in the correct sections and that source URLs are included as footnotes.
Click approve to accept the changes.

What should I see?

The agent should work through the steps methodically. You will see it read the note, perform web searches for each topic, and then propose edits that include new information, a Last Updated date in the frontmatter, and footnote citations.

The key difference from your earlier manual prompting is that you did not need to explain the full process. The skill handled it for you.

Trigger skills with a slash command

You can also type / in an empty chat input to browse and select your skills from a picker. This is faster than typing the full "Use the research-updater skill" prompt.

Agent not following the skill steps?

Make sure you typed the skill name exactly as research-updater in your prompt. If the agent cannot find the skill, it will improvise instead of following your defined process.

If the agent skips steps (for example, jumping to edits without searching first), tell it to slow down and follow each step in the skill explicitly.

Help me get the agent to follow my research-updater skill properly.

Secret mission

Build a Changelog Skill

You have a research-updater skill that refreshes individual notes. But after a long session of updates, how do you remember what changed? In this secret mission, you will create a second agent skill that automatically generates a changelog note summarizing every edit made during a session.

Clean Up Your Resources

Decide whether to keep your resources running, pause them to come back later, or delete them entirely. This project runs entirely locally with a free API key, so there are no ongoing costs.

Resources you used:

Gemini Scribe plugin installed in Obsidian.
Gemini API key from Google AI Studio.
Agent skills stored in gemini-scribe/Skills/ folder.
Agent session files stored in gemini-scribe/Agent-Sessions/.
Modified research notes in your vault.

Keep everything running

No action needed. Choose this if you plan to keep using the research-updater workflow regularly.

Your Gemini Scribe plugin stays active and ready to use.
Your API key remains configured in the plugin settings.
Your agent skills persist in the gemini-scribe/Skills/ folder and are available any time you open Obsidian.

Pause - I'll come back to this later

Disable the plugin to free up resources while keeping your configuration intact.

Open Obsidian Settings and go to Community Plugins.
Toggle off Gemini Scribe to disable it.
Your API key, skills, and session files all stay saved. When you return, toggle the plugin back on and everything works as before.

Delete - I don't want to use this again

Remove the plugin and all associated files from your vault.

Open Obsidian Settings and go to Community Plugins.
Find Gemini Scribe in the list and click the trash icon to uninstall it.
Delete the gemini-scribe/ folder from your vault root. This removes all agent sessions and skills.
To revoke your API key, go to https://aistudio.google.com/apikey and delete the key you created.
Your research notes remain untouched. Any edits the agent made are part of your notes now.

Nice Work!

You now have an AI-powered research assistant living inside your Obsidian vault. Any of your tech research notes can be refreshed with current information in a single command.

You learned how to:

Installed and configured Gemini Scribe with a free Gemini API key for web-grounded research.
Updated a tech research note with current web information using the AI agent's built-in Google Search.
Created a reusable agent skill called research-updater that standardizes your update process every time you trigger it.
Secret Mission: Built an update-changelog skill that generates a summary of all research updates made in a session.

Ready to quiz yourself?