Build an AI Second Brain with Obsidian

Use OpenAI Codex to build a self-maintaining personal knowledge wiki.

Introduction

30 Second Summary

You have years of accumulated knowledge scattered across random files, bookmarks, and your own memory. When you need to recall a decision, a preference, or a lesson learned, finding it feels impossible.

In this project, you will build a self-maintaining AI second brain that organizes your personal knowledge into a searchable, cross-linked wiki. You drop raw notes into a folder, and an AI agent processes them into structured pages while you sleep.

What You'll Build

A personal knowledge management system powered by Obsidian and OpenAI Codex that automatically ingests raw markdown files, creates cross-linked wiki pages, and answers questions grounded in your own accumulated knowledge.

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

A structured Obsidian vault with immutable source files and an AI-maintained wiki that grows smarter over time.
An AGENTS.md configuration that governs how Codex processes, links, and queries your content.
A nightly automation that processes new files without intervention and a queryable knowledge base you can ask questions in plain language.
Secret Mission: Add a Web Clipper ingestion template that auto-captures articles and web content into your raw folder with proper metadata.

Are there any prerequisites?

You need a Codex Pro subscription (included with ChatGPT Plus, Pro, Business, Edu, or Enterprise plans) and Obsidian installed on a Windows device. Both should be set up and ready to go before you begin.

Set Up Your Second Brain Vault

Every piece of knowledge you accumulate (your decisions, your projects, your preferences) lives scattered across random notes, docs, and your memory. A second brain gives that knowledge a permanent, searchable home.

In this step, you will create the folder architecture that makes your second brain work. You will set up the three-layer vault structure, seed it with your first source files, and connect it to Codex so it can read and write directly in your vault.

In this step, get ready to:

Create the vault folder structure in Obsidian.
Seed the vault with personal markdown files.
Connect the vault to the Codex desktop app.

Create the vault folder structure

Your second brain uses the Karpathy three-layer architecture. Raw source files go into one folder. AI-generated wiki pages live in another. Configuration files sit at the vault root.

In Obsidian, create a new vault (or designate an existing vault) to use as your second brain.
Name the vault something meaningful like second-brain or my-wiki.

How do I create a new vault in Obsidian?

When you open Obsidian, the vault switcher appears. Click Create to make a new vault, give it a name, and choose a location on your computer.

If you already have a vault you want to use, open it from the vault switcher instead.

Now create the two core folders. The raw/ folder holds your immutable source files. The wiki/ folder is where Codex will generate and maintain structured pages.

In the File explorer sidebar, click New folder at the top.
Type raw and press Enter.
Click New folder again at the top of the File explorer.
Type wiki and press Enter.

Next, create the three root-level configuration files that Codex will use to manage your knowledge base.

Click New note at the top of the File explorer.
Type AGENTS and press Enter. This creates AGENTS.md at your vault root.
Click New note again. Type index and press Enter.
Click New note one more time. Type log and press Enter.

What is each file for?

AGENTS.md tells Codex how to process, link, and query your knowledge. You will configure this in Step 2.
index.md serves as a master table of contents for all wiki pages.
log.md records every action Codex takes, giving you a full audit trail.

Your File explorer should now show: raw/, wiki/, AGENTS.md, index.md, and log.md.

Don't see all five items?

Make sure the folders and notes were created at the vault root, not inside another folder. If a note ended up in the wrong place, right-click it in the File explorer and select Move file to... to relocate it to the root.

help me fix my vault folder structure

Seed the vault with personal markdown files

Codex needs source material to work with. You will add 2-3 markdown files about yourself into the raw/ folder. These become the first documents Codex will process into wiki pages.

Right-click the raw folder in the File explorer and click New note.
Name it about-me and press Enter.
Write a few paragraphs about yourself: your name, what you do, your current goals, or anything you want your second brain to know.
Create a second note inside raw/ the same way. Name it something like my-projects or my-decisions.
Write content about your current projects, recent decisions, or topics you think about frequently.
Optionally, create a third note in raw/ on any topic that matters to you.

What makes good source material?

Write about things you would want to look up later. Your professional background, project decisions, personal preferences, learning goals, or career plans all work well.

The more specific and personal the content, the more useful your wiki will be. Even a few sentences per file is enough to start.

Your raw/ folder should now contain at least 2 markdown files with personal content.

Connect the vault to the Codex desktop app

With your vault structure in place, you need to point Codex at the vault folder so it can read your files and write new ones. Codex works directly in whatever project folder you select.

Switch to the Codex desktop app.
Click Add new project next to Threads in the sidebar (or press Ctrl+O).
Browse to your vault folder and select it.
Confirm that Local is selected as the thread mode. This ensures Codex works directly in your vault directory.

What does Local mode mean?

Codex offers three thread modes: Local, Worktree, and Cloud. Local means Codex reads and writes files directly in your project folder.

For a personal knowledge vault that is not a Git repository, Local is the correct choice. Codex starts in Agent mode by default, which lets it read files, run commands, and write changes in your project directory.

Verify that your vault folder name appears in the Codex sidebar as the active project.
Check that you can see raw/, wiki/, AGENTS.md, index.md, and log.md in the Codex file explorer.

Codex not showing your files?

Make sure you selected the vault's root folder, not a subfolder like raw/. The project folder should be the one containing all five items.

If you accidentally selected the wrong folder, hover over the project name in the sidebar, click the three dots, and choose Remove. Then add the correct folder.

help me connect my vault to Codex

Your second brain vault is ready. Next up, you will write the AGENTS.md file that teaches Codex exactly how to process, link, and query your knowledge.

Write Your AGENTS.md

Your vault structure is ready with raw source files waiting to be processed. Now you need to tell Codex exactly how to handle your knowledge.

Without instructions, Codex has no idea what to do with the files in your vault. AGENTS.md is a special configuration file that Codex reads before doing any work. It defines the operations Codex can perform, the rules it must follow, and the structure of its output.

In this step, get ready to:

Write three core operations (process-source, answer-query, update-log) in your AGENTS.md file.
Define wiki page structure rules that ensure consistency across all generated pages.
Verify that Codex loads and understands your configuration.

Define the three core agent operations

Codex reads AGENTS.md files before doing any work. This file acts as persistent instructions that shape every interaction.

Switch back to Obsidian and open the AGENTS.md file at the root of your vault.
Replace any placeholder content with the following:

# Second Brain Instructions

## Operations

### 1. Process Source
When asked to process files from `raw/`:
- Read the specified file (or all files in `raw/` if none specified).
- For each distinct concept or topic, create a new page in `wiki/` or update an existing one.
- Cross-link related concepts using `[[page-name]]` wikilink syntax.
- Add a one-line entry for each new or updated page in `index.md`.
- Append a processing record to `log.md` following the update-log format.

### 2. Answer Query
When I ask a question about my knowledge base:
- Read `index.md` to identify relevant wiki pages.
- Read those pages for detailed information.
- Answer with citations to specific wiki pages (e.g., "According to [[page-name]]...").
- Offer to save the synthesized answer as a new wiki page if it produces novel insights.

### 3. Update Log
After every action, append a record to `log.md` with:
- Date (YYYY-MM-DD)
- Action type (processed, queried, updated, created)
- One-line summary of what was done

What does this configuration do?

You have defined three distinct operations that Codex can perform on your vault:

Process Source turns raw markdown files into structured wiki pages with cross-links. This is the ingestion pipeline.
Answer Query lets you ask questions in natural language and get answers grounded in your own knowledge, with citations.
Update Log ensures every action is tracked so you always know what changed and when.

Define wiki page structure rules

The operations tell Codex what to do. Structure rules tell it how the output should look. Without these, every wiki page would have a different format.

In AGENTS.md, add the following content directly below the operations you just wrote:

## Wiki Page Structure

Every page created in `wiki/` must follow these rules:

### Frontmatter
Include YAML frontmatter at the top of every wiki page with these fields:
- title: the page title
- date: the date created or last updated (YYYY-MM-DD)
- source-count: number of source files that contributed to this page

### Linking
- Use `[[page-name]]` internal links to connect related concepts across pages.
- Each page should link to at least one other wiki page where relevant.

### Citations
- Every factual claim must cite its source file (e.g., "Source: raw/about-me.md").
- Do not generate information that is not grounded in a source file.

### Contradictions
- If new information conflicts with an existing wiki page, add a contradiction flag noting the conflict and both sources.
- Never silently overwrite existing information.

### Naming
- Filenames: lowercase with hyphens (e.g., `my-priorities.md`).
- One core concept per page.

Why are these rules important?

The YAML frontmatter gives every page consistent metadata that Codex can read programmatically. The [[wikilink]] syntax creates clickable connections between pages inside Obsidian.

The contradiction flag is especially valuable. As your knowledge base grows, you will inevitably add information that conflicts with something written months earlier. Instead of silently overwriting, Codex preserves both versions so you can decide which is correct.

Save AGENTS.md in Obsidian.

File not saving?

Make sure you are editing the AGENTS.md file at the vault root, not inside the raw/ or wiki/ folders. Codex looks for AGENTS.md starting at the project root and walking down the directory tree.

help me find where AGENTS.md should be located in my vault

✔️ Awesome, I've got everything!

Great. Double check you have saved AGENTS.md and move on to testing.

ⓧ I'd like to double check the full code

Here is the complete AGENTS.md file with both sections combined:

# Second Brain Instructions

## Operations

### 1. Process Source
When asked to process files from `raw/`:
- Read the specified file (or all files in `raw/` if none specified).
- For each distinct concept or topic, create a new page in `wiki/` or update an existing one.
- Cross-link related concepts using `[[page-name]]` wikilink syntax.
- Add a one-line entry for each new or updated page in `index.md`.
- Append a processing record to `log.md` following the update-log format.

### 2. Answer Query
When I ask a question about my knowledge base:
- Read `index.md` to identify relevant wiki pages.
- Read those pages for detailed information.
- Answer with citations to specific wiki pages (e.g., "According to [[page-name]]...").
- Offer to save the synthesized answer as a new wiki page if it produces novel insights.

### 3. Update Log
After every action, append a record to `log.md` with:
- Date (YYYY-MM-DD)
- Action type (processed, queried, updated, created)
- One-line summary of what was done

## Wiki Page Structure

Every page created in `wiki/` must follow these rules:

### Frontmatter
Include YAML frontmatter at the top of every wiki page with these fields:
- title: the page title
- date: the date created or last updated (YYYY-MM-DD)
- source-count: number of source files that contributed to this page

### Linking
- Use `[[page-name]]` internal links to connect related concepts across pages.
- Each page should link to at least one other wiki page where relevant.

### Citations
- Every factual claim must cite its source file (e.g., "Source: raw/about-me.md").
- Do not generate information that is not grounded in a source file.

### Contradictions
- If new information conflicts with an existing wiki page, add a contradiction flag noting the conflict and both sources.
- Never silently overwrite existing information.

### Naming
- Filenames: lowercase with hyphens (e.g., `my-priorities.md`).
- One core concept per page.

Test the configuration in Codex

Writing the file is only half the job. You need to confirm that Codex actually reads and understands your instructions.

Switch back to the Codex app with your vault project open.
In the thread composer, type /status and press Enter.

What does /status show?

The /status command shows the thread ID, context usage, and rate limits. It confirms that Codex is operating in your vault directory and has loaded your project files.

Verify that the output shows your vault as the active project directory.
Now ask Codex to confirm it understands your instructions by typing this message in the composer:

Summarize the instructions in AGENTS.md. What operations can you perform and what rules must you follow?

Press Enter to send the message.
Read Codex's response. It should describe all three operations (process-source, answer-query, update-log) and the wiki page structure rules (frontmatter, linking, citations, contradictions, naming).

Refine if needed

If Codex's summary misses something or misinterprets a rule, edit AGENTS.md in Obsidian and ask Codex to re-read it. The file is capped at 32 KiB by default, so you have plenty of room to add detail.

Codex does not mention your operations?

Confirm the file is named exactly AGENTS.md (uppercase) at the vault root. Codex checks for AGENTS.md first by default. A lowercase agents.md may not be picked up as the primary instruction file.

Also make sure you saved the file in Obsidian before asking Codex. Unsaved changes are not visible to the Codex agent.

help me troubleshoot why Codex is not reading my AGENTS.md file

Your second brain now has a brain of its own. Next up, you will put these instructions to work by running your first ingestion and watching Codex transform raw files into a cross-linked wiki.

Run Your First Ingestion

Your vault is structured, your source files are in place, and AGENTS.md is loaded. Now it is time to see the system work.

In this step, you will prompt Codex to process your raw files, review what it produces, and refine the instructions if anything is off. This is the moment your second brain starts thinking.

In this step, get ready to:

Trigger the full ingestion pipeline by prompting Codex.
Review the generated wiki pages, index, and log in Obsidian.
Iterate on AGENTS.md to refine how Codex structures your knowledge.

Prompt Codex to process your raw files

The process-source operation you defined in AGENTS.md tells Codex exactly what to do with each file in raw/. You just need to tell it to start.

In the Codex app, type the following prompt into the message input:

Process all the files in the raw folder.

Press Enter to send the prompt.

Codex reads AGENTS.md, locates the process-source operation, and executes the full pipeline on each file in raw/. This may take a minute depending on how many files you have.

What is Codex doing right now?

Codex is running the process-source operation for each file in your raw/ folder. For each file, it creates or updates wiki pages in wiki/, cross-links related concepts using [[wikilinks]], adds entries to index.md, and appends processing records to log.md.

Because Codex runs in Agent mode by default, it can read files, run commands, and write changes directly in your vault directory.

Wait for Codex to finish processing. You will see it report actions for each file it reads and each wiki page it creates.

Codex not processing all files?

If Codex only processes one file or asks for clarification, confirm that your AGENTS.md clearly defines the process-source operation with instructions to iterate over all files in raw/. You can also try rephrasing: "Process each file in the raw/ folder one by one, following the process-source operation."

help me troubleshoot why Codex is not processing all my raw files

Review the output in Obsidian

Codex has written new files into your vault. Now you need to verify the output matches your expectations for structure, linking, and citations.

Switch to Obsidian and open the wiki/ folder in the sidebar.
Click on one of the generated wiki pages to open it.

Each wiki page should have three things. YAML frontmatter at the top with title, date, and source-count fields. Internal [[links]] to related concepts. Citations back to the source files in raw/.

Verify the page has YAML frontmatter between --- markers at the very top.
Look for [[page-name]] links in the body text that connect to other wiki pages.
Confirm that factual claims include citations referencing which file in raw/ the information came from.

Open index.md at the vault root. Verify it now lists all new wiki pages with one-line descriptions.
Open log.md at the vault root. Confirm it shows timestamped processing records for each file Codex handled.

What should I see in the graph view?

Open the graph view in Obsidian by pressing Ctrl+G. You should see nodes for each wiki page connected by lines representing the [[internal links]] Codex created.

The more connections between pages, the richer your knowledge graph becomes. This is how your second brain grows smarter over time.

Wiki pages missing frontmatter or links?

If your wiki pages are missing YAML frontmatter or [[links]], your AGENTS.md may not be specific enough about the required output format. Do not worry. The next substep covers exactly how to fix this through iteration.

help me figure out why my wiki pages are missing structure

Iterate on AGENTS.md to refine the output

Your first ingestion is rarely perfect. Codex may name pages differently than you expected, skip cross-links, or add too little detail. This is normal and fixable.

The power of AGENTS.md is that you can refine it and every future session inherits the correction. This creates a feedback loop that makes the system smarter with each iteration.

Identify what you want to improve. Common issues include wrong page naming conventions, missing links between related pages, or insufficient detail in the wiki entries.
Return to the Codex app and tell it what to fix. For example:

The wiki pages you created are missing links between related concepts. Update AGENTS.md to require at least 2 [[internal links]] per wiki page, connecting to other concepts in the wiki. Then reprocess the raw files.

What makes a good refinement prompt?

Be specific about what is wrong and what you want instead. "The pages need more links" is vague. "Each wiki page should contain at least 2 [[internal links]] to other wiki pages" gives Codex a concrete rule to encode in AGENTS.md.

Ask Codex to update AGENTS.md with the correction so future ingestions use the improved instructions automatically.

After Codex updates AGENTS.md and reprocesses the files, switch back to Obsidian and review the updated wiki pages.
Repeat this cycle until your wiki pages have the structure, detail, and linking density you want.

You do not need perfection here

One or two rounds of refinement is enough for now. You will continue improving AGENTS.md over time as you add more content and notice new patterns. The system is designed to evolve with you.

Confirm your final state: wiki/ contains at least one page per source file, index.md lists all pages, and log.md shows timestamped records.

Your vault is no longer empty. You have a working knowledge base with structured pages, cross-links, and a searchable index. Next up, you will query this knowledge base in natural language and get answers grounded in your own sources.

Query Your Knowledge Base

Your wiki is populated with structured, cross-linked pages built from your own source material. Now it's time to use it.

In this step, you'll ask Codex natural language questions that require synthesizing information across multiple wiki pages. You'll also save useful answers back into the wiki and test how the system handles conflicting information.

In this step, get ready to:

Ask Codex a question that requires synthesizing information from your wiki.
Save a synthesized answer as a new wiki page.
Test a contradiction scenario and verify it gets flagged.

Ask a natural language question

Your wiki pages contain structured knowledge extracted from your raw files. Codex can read the index, locate relevant pages, and synthesize an answer that draws from multiple sources.

In the Codex chat, type a question that requires pulling information from more than one wiki page. Use a prompt like this:

Based on my wiki, what are my main priorities right now?

Alternatively, ask a topic-specific question like: "What decisions have I made about your topic?"
Press Enter and wait for Codex to respond.

What is Codex doing behind the scenes?

Codex follows the answer-query operation you defined in AGENTS.md. It reads index.md to find relevant wiki pages, reads those pages, then synthesizes an answer with citations pointing back to specific pages in your vault.

Codex responds with an answer that cites specific wiki pages using [[wikilinks]]. You should see references to pages in your wiki/ folder throughout the answer.

Codex not citing specific pages?

If Codex gives a generic answer without page citations, remind it to follow the answer-query operation in AGENTS.md. Try rephrasing your question to be more specific about the topic area.

Check that your index.md file has entries with descriptive one-line summaries. If the index is sparse, Codex may not know which pages to look at.

Help me get Codex to cite wiki pages in its answers

Save an answer as a new wiki page

When Codex produces a useful synthesis, that answer itself becomes valuable knowledge. Saving it as a wiki page means your knowledge base grows beyond just ingested sources. Your own thinking becomes part of the searchable system.

After receiving a useful answer, ask Codex to save it as a wiki page by sending this prompt:

Save this answer as a new wiki page in the wiki/ folder. Follow the wiki page structure rules in AGENTS.md - include YAML frontmatter, cross-link to related existing pages, and update index.md and log.md.

Why save answers back into the wiki?

Raw files capture what you know. Wiki pages capture what Codex extracts. Saved answers capture what you learn by querying. Over time, these synthesized pages become the most valuable part of your knowledge base because they connect ideas across multiple sources.

In Obsidian, open the wiki/ folder and verify the new page exists.
Check that the page has YAML frontmatter, [[wikilinks]] to related pages, and citations to the source wiki pages it synthesized from.
Open index.md and confirm the new page is listed with a one-line description.
Open log.md and confirm a new entry records the answer-query action.

Page not appearing in the wiki folder?

Make sure you explicitly asked Codex to save the page in wiki/. Without specifying the folder, Codex may save it at the vault root.

If the page exists but has no frontmatter or links, remind Codex to follow the wiki page structure rules in AGENTS.md. You can ask it to reformat the page.

Help me get Codex to save answers as properly formatted wiki pages

Test a contradiction scenario

Real knowledge evolves. Your priorities shift, decisions get revised, and facts change. A good knowledge system flags when new information conflicts with what it already knows instead of silently overwriting.

Think of a piece of information in your existing wiki that you can contradict. For example, if a wiki page says your top priority is learning Python, you'll create a file that says your top priority changed to something else.
In Obsidian, create a new markdown file inside raw/ that contains information directly conflicting with something already in your wiki.

Make the contradiction obvious

Use clear language like "As of today, my priority has changed from X to Y" or "I no longer believe Z." The more explicit the contradiction, the easier it is to verify that Codex flags it correctly.

In Codex, prompt it to process the new file:

Process all unprocessed files in the raw folder.

Wait for Codex to finish processing.
In Obsidian, open the wiki page that relates to the conflicting information. Look for a contradiction flag in the page content. Your AGENTS.md instructs Codex to flag contradictions rather than silently overwriting.
Open log.md and verify that the processing record for this file is logged.

What does a contradiction flag look like?

The exact format depends on how Codex interprets your AGENTS.md rules. It might add a callout block, a note in the frontmatter, or a dedicated section. The key is that the old information is preserved alongside the new, with a clear note that they conflict.

Codex silently overwrote instead of flagging?

If the old information disappeared without a flag, your AGENTS.md contradiction rule may need to be more explicit. Tell Codex what went wrong and ask it to update AGENTS.md with a stronger instruction about preserving old information and marking the conflict.

Then reprocess the file to see if the updated rule fixes the behavior. This iterative refinement is exactly how you tune the system over time.

Help me get Codex to flag contradictions instead of overwriting

Your knowledge base is alive. It answers questions, grows from your own thinking, and handles conflicting information gracefully. Next up, you'll set up a nightly automation so new content gets processed while you sleep.

Set Up Nightly Automation

Your knowledge base can now answer questions, save synthesized answers, and flag contradictions. The one remaining bottleneck is you.

Every time you drop a new file into raw/, you still have to manually prompt Codex to process it. In this step, you will set up a nightly automation that does this for you while you sleep.

In this step, get ready to:

Create a recurring automation in the Codex Automations panel.
Schedule it to run daily at a time when you are not working.
Test the automation by dropping a new file and confirming it processes without manual intervention.

Open the Automations panel and create your automation

Codex Automations let you schedule recurring tasks that run on a cadence you define. You choose the project, prompt, cadence, and execution environment.

In the Codex app, click Automations in the sidebar.
Click to create a new automation.
Paste the following prompt into the automation prompt field:

If there are any unprocessed files in the raw directory, please process them now following the process-source operation in AGENTS.md.

What does this prompt do?

This prompt tells Codex to check the raw/ folder every time the automation runs. If new files exist that have not been processed yet, Codex follows the process-source operation you defined in AGENTS.md.

Because Codex reads AGENTS.md before doing any work, it already knows exactly how to create wiki pages, cross-link concepts, update the index, and log every action.

Configure the schedule and project

Your vault is not a Git repository, so Codex runs the automation directly in the project directory. This means it reads and writes your actual vault files.

Set the cadence to daily.
Choose a time when you are not working (for example, 12:50 AM).
Select your second brain vault as the project.
Ensure the execution environment is set to your local environment (not a Git worktree).

Why run at night?

Running the automation overnight means Codex processes new files without interrupting your work. By morning, any notes you dropped into raw/ the previous day are already organized into wiki pages.

Codex automations work best when your computer is awake and the Codex app is running. If your machine sleeps overnight, the automation will run the next time both conditions are met.

Don't see the Automations panel?

Make sure you are using the Codex desktop app (not the CLI or IDE extension). The Automations panel lives in the app sidebar.

If the sidebar is collapsed, expand it by clicking the sidebar toggle at the top of the Codex window.

Help me find the Automations panel in the Codex app.

Test the automation with a new file

The best way to confirm the automation works is to give it something to process and watch the results appear in your vault.

Open your raw/ folder in Obsidian.
Create a new markdown file with some personal knowledge (for example, a note about a recent decision or project update).
Save the file inside raw/.
Return to the Codex app and trigger the automation manually, or wait for the next scheduled run.

After the automation finishes, open the wiki/ folder in Obsidian. You should see new wiki page(s) generated from your test file.
Open index.md. Confirm it now lists the new wiki page(s) with one-line descriptions.
Open log.md. Confirm a new timestamped processing record appears for the automation run.

What should I see?

Your wiki/ folder now has at least one new page that was created entirely by the automation. The page should have YAML frontmatter, [[internal links]] to related concepts, and citations back to your test file in raw/.

The key difference from earlier steps is that you did not manually prompt Codex. The automation handled the entire pipeline on its own.

Automation ran but nothing appeared in wiki/?

Check log.md for a recent entry. If no entry exists, the automation may not have triggered yet. Give it a few minutes or trigger it manually again.

If log.md shows a processing record but wiki/ is empty, review your AGENTS.md to ensure the process-source operation specifies the correct output folder.

My Codex automation ran but no wiki pages were created. Help me debug.

Secret mission

Add Web Clipper Ingestion

Your second brain currently relies on you manually creating markdown files. What if you could capture any web article directly into your ingestion pipeline with a single click? In this secret mission, you'll install the Obsidian Web Clipper extension and wire it up so clipped articles land in your raw/ folder, ready for Codex to process automatically overnight.

Clean Up Your Resources

This project runs entirely locally and uses your existing Codex Pro subscription, so there are no ongoing costs. Decide whether to keep your resources running, pause them to come back later, or delete them entirely.

Resources you used:

Obsidian vault with AI-maintained wiki (local files on your machine).
Codex nightly automation (runs on your existing Codex Pro subscription, no additional cost).
Obsidian Web Clipper extension in Microsoft Edge (free).

Keep everything running

No action needed. Choose this if you plan to keep using your second brain daily.

The nightly automation continues processing new files indefinitely at no extra cost beyond your existing Codex Pro plan.
Any new markdown files you drop into raw/ or clip with Web Clipper will be processed automatically on the next scheduled run.
Your vault stays on your local machine. No cloud storage or external service holds your data.

Pause - I'll come back to this later

Disable the automation to stop nightly processing. Your vault and all wiki pages remain intact on your machine.

Open the Codex app and navigate to the Automations panel.
Find your nightly processing automation and disable it.
Your vault folder, wiki pages, and Web Clipper extension all remain in place. You can re-enable the automation anytime to resume processing.

Delete - I don't want to use this again

Remove all project resources and start fresh. This deletes your vault, wiki pages, and all configuration.

Delete the automation: open the Codex app, go to the Automations panel, find your nightly automation, and delete it.
Delete the project from Codex: in the Codex app, remove the second brain project so Codex no longer references your vault folder.
Remove the Obsidian Web Clipper extension: in Microsoft Edge, right-click the Web Clipper icon next to the address bar, select Remove from Microsoft Edge, then click Remove to confirm.
Delete the vault folder from your machine. Open File Explorer, navigate to the folder where you created your vault, right-click the vault folder, and select Delete. This removes all files including raw/, wiki/, AGENTS.md, index.md, and log.md.

Nice Work!

You just built a self-maintaining AI second brain powered by Obsidian and OpenAI Codex. Your personal knowledge system now ingests raw notes, generates cross-linked wiki pages, and answers questions grounded in your own accumulated thinking.

You've learned how to:

Built a three-layer knowledge vault (raw sources, AI-maintained wiki, root config files) following the Karpathy LLM Wiki architecture.
Wrote an AGENTS.md configuration that governs how Codex processes, links, and queries your personal knowledge.
Set up a nightly automation that processes new content while you sleep and grows your knowledge base without manual intervention.
Secret Mission: Added Web Clipper ingestion to capture web articles directly into your raw folder for automatic processing into cross-linked wiki pages.

Ready to quiz yourself?