Your AI agent's files are dead weight until you connect them to Obsidian
The core argument is a distinction between dead files and living files. A dead file is any document no AI agent can access or act on—whether it sits on a local hard drive, Google Drive, or even inside an AI tool with no visual interface. A living file is a markdown document an agent can load as a skill, memory, or context on demand. The host's thesis is that most people's AI outputs—research reports, skill definitions, notes—are effectively dead the moment they're created.
The proposed solution is linking the Hermes agent (an open-source agentic framework approaching 200k GitHub stars) to an Obsidian vault, synced in near-real-time across a MacBook, phone, and a Hostinger VPS via Obsidian Sync ($4–5/month) or the free Syncthing alternative. The host demonstrates the full setup live: installing Obsidian Headless on the VPS via a Pi agent prompt, initializing encrypted sync, and confirming a file created on the VPS appears in the desktop app in under half a second.
Hermes loads context in two layers—a small always-on "soul" file with core user preferences, and skills/notes pulled on demand—so the vault doesn't bloat the context window. The host has 185 skills, half custom, all editable directly in Obsidian. A live demo uses the `/go` command to autonomously scrape YouTube for the 50 highest-viewed Claude Code videos, fetch transcripts, and save each as a markdown file in the vault—a task the host estimates would take a human 3–4 hours.
The bottom line is practical and opinionated: running a VPS, keeping files in an agent-accessible vault, and using tools like CMAX with SSH is framed as the defining productivity advantage of 2026. The host is blunt that failure to see use cases reflects a lack of imagination or setup rather than any limitation of current AI.
The host walks through connecting the Hermes AI agent to an Obsidian vault synced across devices and a VPS, turning static skill files into 'living files' any agent can use instantly. The setup is framed as a major productivity unlock for 2026, letting a single user orchestrate multiple AI agents with full visibility and control.
Having the right agentic setup—such as CMAX with a PI agent on the left and SSH into a VPS on the right—lets users outcompete people doing everything manually, making it the biggest productivity cheat code in 2026.
Everyone should absolutely be running their own virtual private server in 2026
Connecting Hermes agent to Obsidian makes the human user more in charge as an orchestrator because they can see visual graphs, connectors, patterns, clear folders, and easily editable markdown files.
The host currently has 185 skills in Hermes, roughly half pre-built and half custom.
Editing an Obsidian note updates the skill file and the AI agent will use the new version of that file instantly.
Living files are just markdown files that can be used by any AI agent, meaning an Obsidian vault can be loaded into Cursor, Open Code, Claude Code, Codex, or any other AI agent instantly.
The Hermes agent GitHub repository is almost at 200,000 stars.
GLM 5.2 from ZhipuAI is probably the best open source model right now and is way more cost efficient than Opus, being the first open source model that doesn't feel like it's behind closed source models.
Obsidian Sync costs $4 a month and provides encrypted syncing so even Obsidian staff cannot access your notes; Syncthing is an open source free alternative requiring more setup.
The host is privately building a custom AI inference provider with complete token controls, observability, graphs, per-model and per-API-key limits, and may release it publicly depending on interest.
Passwords pasted directly into the terminal do not go to Anthropic and stay private on the VPS
A file created on the VPS synced and appeared in Obsidian on the MacBook
Sync speed between VPS and MacBook Obsidian was less than half a second
Hermes agent skills are pre-packaged as .md files and can be edited directly in Obsidian
The /go feature works until a verifiable, quantifiable end outcome is achieved, regardless of whether that takes 10 minutes, 30 minutes, or 5 hours
Codex probably has the best /go feature; Hermes agent has the second best, but Hermes is a general agent not optimized for coding only
Finding the 50 highest-views videos on YouTube about Claude Code and saving them manually would take at least 3–4 hours of real human work
Most people never look at their skills in Hermes at all, meaning they have skills unused for months and the skills they do use are outdated and not relevant.
When an AI agent creates a file somewhere without a visual interface, users either need to remember it or they forget it was created.
Deep research run in ChatGPT that is not saved into an easily accessible living file on a VPS is useless because the output does not continue working for you.
Loading an entire Obsidian vault into the agent's memory would be very inefficient, cost a lot of tokens, and bloat the context window.
Before this setup, users did not have good observability on what the agent was doing
A 'dead file' is any file that no AI agent can access, use as context, compare to other files, or perform research on—regardless of whether it is on a disconnected hard drive, local computer, or Google Drive.
A 'living file' is a file that can be used by an AI agent as a skill, reference, memory, context, or part of a prompt to do tasks more efficiently.
Hermes agent loads context in two parts: a small always-on system prompt (the 'soul' file) containing core memory like the user's name and response preferences, and skills/notes that are only loaded on demand when relevant to avoid bloating the context window.
Encrypt private data in an Obsidian vault, sync it across phone, MacBook, and VPS, and let Hermes agent manage the second brain
Set up Hermes agent on a Hostinger KVM2 VPS, which the host claims is sufficient to run Hermes agent, Open Claude, Py agent, Agent Zero, and Codex CLI simultaneously.
Use Pi agent running in CMAX to manage the VPS and install Hermes agent via the official quick install command from GitHub, rather than manually running terminal commands.
Purchase the monthly $5 plan for Obsidian Sync
Initialize Obsidian Sync by choosing remote vault, creating new vault in remote 02 region with automatic encryption and a secure password
Set up a proper password manager before creating vault credentials
Install Obsidian Headless on the VPS by pasting a full setup prompt into the Pi agent
Use /go to instruct Hermes agent to scrape YouTube for the 50 highest-viewed videos about Claude Code, fetch their transcripts as raw txt files, and save them into the Obsidian vault as separate markdown files
“Having the right agentic setup is really the biggest productivity cheat code in 2026.”
“Over 60% of you are not subscribed.”
“if you don't see the use cases, it really is a poverty of imagination. You probably just lack creativity. It's not that the AI isn't useful enough. You just don't have the right setup or you haven't figured out what the use cases are for you.”
Shared with Earmark · earmark-ai.com