@jacobcole / IdeaFlow Obsidian LLM Wiki › hermes-agent-obsidian-ideaflow-second-brain-tutorial
Public View
Suggest
Download this page (.md) Download entire wiki (.zip)
Clone entire wiki

Hermes Agent + Obsidian/IdeaFlow Second Brain Tutorial

A beginner-friendly path for setting up a personal AI agent with a durable “second brain”: Hermes Agent does the work, IdeaFlow captures living thoughts and heartbeat triggers, and Obsidian / Markdown stores a browsable LLM Wiki you can keep improving over time.

[!info] Essential links

What you are building

By the end, you should have:

  1. Hermes Agent installed locally or on a remote machine.
  2. A Markdown LLM Wiki folder that acts as your durable knowledge base.
  3. Obsidian or another Markdown reader to browse and edit the wiki.
  4. IdeaFlow / Thoughtstream heartbeat hooks that let notes trigger useful agent actions.
  5. Hermes skills that teach the agent how to find and maintain your wiki.

Think of this as three layers:

Capture layer:      IdeaFlow / Thoughtstream notes, meetings, #tags, heartbeat
Agent layer:        Hermes Agent + skills + cron / heartbeat jobs
Knowledge layer:    Markdown LLM Wiki opened in Obsidian / WikiHub / Markdown tools

Step 1 — Install Hermes Agent locally

For beginners on macOS or Windows, start with Hermes Desktop:

Hermes Desktop installs both the desktop app and the hermes command-line tool.

If you prefer command-line only on Linux, macOS, WSL2, or Termux:

curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash

Then run the simplest setup path:

hermes setup --portal
hermes doctor

hermes setup --portal is currently the fastest beginner path because one OAuth setup can configure a model provider plus built-in tool gateway features like web search, browser, image generation, and TTS.

Step 2 — Optional remote/server setup in two steps

If you want your agent to run somewhere that is not your laptop, use either a normal remote server over SSH or a managed cloud workspace such as Daytona. If you have access to cortex.ideaflow.app, that can be a good free/low-friction place to start experimenting with a hosted workspace.

Option A: install Hermes directly on the remote server

  1. SSH into the server and install Hermes:
ssh ubuntu@your-server
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup --portal
hermes doctor
  1. Keep it reachable from your phone or desktop using the gateway when ready:
hermes gateway setup
hermes gateway run

Use this when you want Hermes itself to live on the server.

Later: local Hermes + SSH remote backend

There is also a useful advanced setup where Hermes Desktop stays local, but terminal work executes on a remote server over SSH. We are leaving that out of the beginner path for now to keep the tutorial simple. Consider bringing it back later for users who already have a VPS and want local Desktop UX with remote execution.

Option B: Daytona cloud workspace backend

Daytona is a managed cloud workspace backend. It is useful when you want persistent cloud sandboxes that can stop/resume instead of running all the time.

  1. Add a Daytona API key and enable the backend:
hermes config set DAYTONA_API_KEY your_daytona_key_here
hermes config set terminal.backend daytona
hermes config set terminal.container_persistent true
  1. Verify it:
hermes doctor
hermes chat -q "Create a hello.txt file, print pwd, and confirm this is running in Daytona" --toolsets terminal,file

Docs: Hermes terminal backend configuration.

Step 3 — Create your personal LLM Wiki folder

An LLM Wiki is a Markdown knowledge base that the agent maintains over time. The important files are:

wiki/
├── SCHEMA.md       # rules, domain, tags, page conventions
├── index.md        # map of every page and one-line summaries
├── log.md          # append-only history of wiki changes
├── raw/            # immutable source material
├── entities/       # people, orgs, products, projects
├── concepts/       # reusable ideas and topics
├── comparisons/    # side-by-side analyses
└── queries/        # useful answers worth saving

Create a folder:

mkdir -p ~/wiki/{raw/articles,raw/transcripts,raw/assets,entities,concepts,comparisons,queries}

Then ask Hermes:

Create a new LLM Wiki in ~/wiki. The domain is my personal second brain: projects, people, meetings, IdeaFlow notes, useful decisions, and recurring workflows. Use the llm-wiki skill conventions. Create SCHEMA.md, index.md, and log.md.

The AGI House LLM Wiki is a useful example of the end state: AGI House LLM Wiki.

Step 4 — Open the wiki in Markdown tools

Recommended tools:

  • Obsidian — best for browsing, graph view, wikilinks, backlinks, and manual editing.

  • OpenMarkdownReader v1.0.15 — lightweight Markdown reading tool.

  • WikiHub — useful for publishing selected pages privately or publicly.

In Obsidian:

  1. Open ~/wiki as a vault.
  2. Keep [[wikilinks]] enabled.
  3. Set attachments to raw/assets/.
  4. Optional: install Dataview for frontmatter-powered tables.

Step 5 — Tell Hermes where the wiki lives

Add the wiki path to Hermes’ environment:

hermes config set WIKI_PATH ~/wiki
hermes config set OBSIDIAN_VAULT_PATH ~/wiki

Then install or load the useful wiki skills:

hermes skills install llm-wiki
# When the IdeaFlow skill PR lands upstream, install it from the skills catalog.
# Until then, inspect/install from the public SKILL.md URL:
hermes skills inspect https://raw.githubusercontent.com/tmad4000/hermes-agent/feat/ideaflow-obsidian-llm-wiki/skills/research/ideaflow-obsidian-llm-wiki/SKILL.md

In a Hermes chat, you can also say:

Use the llm-wiki skill. My wiki is at ~/wiki. First read SCHEMA.md, index.md, and the last 30 lines of log.md before making changes.

The most important maintenance skill is llm-wiki: it teaches Hermes to orient on SCHEMA.md, index.md, and log.md; ingest sources into raw/; update entity/concept pages; maintain wikilinks; and lint the wiki for broken links, missing frontmatter, stale pages, and orphan notes.

Use ideaflow-obsidian-llm-wiki when you want IdeaFlow / Thoughtstream to feed that same wiki through heartbeat-driven imports.

Step 6 — Connect IdeaFlow / Thoughtstream heartbeat ideas

The IdeaFlow workflow is:

  1. IdeaFlow captures live thoughts, tasks, tags, and meeting notes.
  2. A heartbeat job checks for new or changed notes.
  3. Hermes imports durable notes into raw/ideaflow/ or raw/transcripts/.
  4. Hermes updates the LLM Wiki pages, index, and log.
  5. Obsidian / WikiHub become the readable surface for the compiled knowledge.

Use the skill:

Example heartbeat prompt:

Every heartbeat, check IdeaFlow for notes changed since the last run. If a note has #llmwiki, import it into my LLM Wiki. If a note has #live-meeting, append concise running bullets to the note and save durable decisions to raw/transcripts/. If a note has #defer1mo, turn it into a deferred follow-up task for one month from the note date. Always update index.md and log.md.

Example heartbeat use cases from the Tejas / IdeaFlow meeting

Source note: Otter transcript titled “AI Integration Workshop” from the recent meeting set. The transcript discussed several concrete IdeaFlow + Hermes heartbeat ideas:

  • Read recent IdeaFlow notes: ask Hermes to read a few recent notes from IdeaFlow, summarize them, and identify follow-ups.

  • Keyword-triggered agents: if a note contains a certain keyword or tag, the heartbeat can invoke an agent action.

  • Inline agent responses: a note can say @agent research this, @agent file a bug report, or @agent enrich this, and Hermes can respond by appending output back into the note or wiki.

  • Live meeting notes: if a note has #meeting or #live-meeting, the heartbeat can append semi-real-time bullets like “key things to remember from this meeting.”

  • Faster heartbeat during live meetings: once #live-meeting is detected, temporarily run the heartbeat more often so useful notes appear while the meeting is still happening.

  • Deferred tasks: tags like #defer1mo can become future reminders or tasks, e.g. “cancel subscription one month from this note’s date.”

  • Cross-channel follow-ups: Hermes can transform meeting or IdeaFlow notes into follow-ups for email, docs, GitHub issues, bug reports, or personal task lists.

A beginner-friendly tag taxonomy might be:

#llmwiki       import this into the durable wiki
#meeting      summarize after the meeting
#live-meeting append running bullets during the meeting
#defer1mo     create a one-month deferred follow-up
#agent        explicit instruction for Hermes
#bug          file or draft a bug report
#research     do background research and add sources
#decision     preserve as a decision record

Step 7 — First “agent with second brain” test

Try this end-to-end beginner test:

  1. Create a note in IdeaFlow or Obsidian:
# First second-brain test #llmwiki #agent

I want Hermes to remember that my second brain is a Markdown LLM Wiki in ~/wiki.
The wiki should track projects, meetings, people, ideas, and deferred follow-ups.

@agent Add this as a concept page and update the index/log.
  1. Tell Hermes:
Use the llm-wiki skill. Process my first second-brain test note into ~/wiki. Create or update the right concept page, add wikilinks, update index.md, and append log.md.
  1. Open Obsidian and verify:

  • A new concept/entity page exists.

  • index.md links to it.

  • log.md records the change.

  • The page has at least two useful wikilinks or placeholders for related concepts.

Step 8 — Keep it alive with scheduled maintenance

Once the basics work, schedule maintenance:

Create a Hermes cron job that runs every day at 8am. It should use the llm-wiki skill, read ~/wiki/SCHEMA.md, ~/wiki/index.md, and the last 30 lines of ~/wiki/log.md, then report broken links, orphan pages, stale pages, and suggested next sources to ingest.

For IdeaFlow heartbeat sync, schedule a shorter interval only if the source is safe and rate limits are understood:

Create a Hermes cron job every 30 minutes that uses ideaflow-obsidian-llm-wiki. It should check for changed IdeaFlow notes, import notes tagged #llmwiki or #meeting into ~/wiki, and append a concise summary of changes to log.md. Do not expose tokens or raw private notes outside the private wiki.

Safety defaults

  • Keep private notes private by default.

  • Store API keys only in ~/.hermes/.env, Keychain, or provider auth stores — never in wiki pages.

  • Keep raw sources in raw/ immutable; edits and synthesis go in entity/concept pages.

  • Ask before publishing pages publicly.

  • Use remote, Docker, SSH, Daytona, or Cortex-style sandboxes for risky code execution.

The shortest possible recipe

# 1. Install Hermes
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup --portal

# 2. Create wiki folder
mkdir -p ~/wiki/{raw/articles,raw/transcripts,raw/assets,entities,concepts,comparisons,queries}
hermes config set WIKI_PATH ~/wiki
hermes config set OBSIDIAN_VAULT_PATH ~/wiki

# 3. Ask Hermes to initialize it
hermes chat -q "Use llm-wiki. Create a personal second-brain LLM Wiki in ~/wiki for projects, meetings, people, IdeaFlow notes, and recurring workflows."

# 4. Open it
# Download Obsidian: https://obsidian.md/

That gives you a simple agent + second brain loop: capture in IdeaFlow, maintain with Hermes, browse in Obsidian, and publish selected knowledge through WikiHub.

[[curator]]
I'm the Curator. I can help you navigate, organize, and curate this wiki. What would you like to do?