Obsidian + Hermes Agent: A Second Brain That Talks Back
Avi Chawla's post on Hermes Agent integrations hit a nerve this week, pulling 1,129 likes and 164 retweets in under 24 hours. The top item on his list: the Obsidian integration, described as "a Karpathy-style second brain, but one that talks back."
The post resonated because it articulated something people have wanted from their knowledge bases for years. Obsidian is already the go-to tool for building a linked knowledge graph - flat markdown files, bidirectional links, graph view. But it has always been a passive system. You write. You search. You connect. The system sits there until you interact with it.
The Hermes Agent Obsidian plugin changes that equation.
What the integration actually does
The plugin gives Hermes filesystem-level access to an Obsidian vault. It can read any note, search across the entire vault by filename or content, create new notes, append to existing ones, and make targeted edits. The agent also understands Obsidian's [[wikilink]] syntax and can create cross-references between notes.
The integration is filesystem-first. Hermes reads .md files directly from the vault directory. There is no API layer, no sync protocol, no database. If the file is in the vault, the agent can access it.
Here is a breakdown of the available operations:
| Operation | Tool Used | Description |
|---|---|---|
| Read a note | read_file |
Read any markdown note by absolute path |
| List notes | search_files |
Glob match *.md across vault directories |
| Search contents | search_files |
Regex content search across all notes |
| Create note | write_file |
Write new markdown note to vault path |
| Append content | patch |
Anchored append after headings or known context |
| Targeted edit | patch |
Find-and-replace in specific note sections |
| Wikilink creation | write_file |
Generate [[Note Name]] cross-references |
From passive to active
A standard Obsidian vault is a graph you explore. You navigate links. You open the graph view. You search. The agent operates differently - it traverses the graph automatically as part of its reasoning.
When Hermes loads the Obsidian skill, it gains a working directory full of interconnected facts. Notes become function arguments. Backlinks become context. The agent can follow a chain of [[wikilinks]] across multiple notes to assemble a complete picture before responding.
This is what Chawla means by "reasons over it." A typical LLM receives a static context window. An agent with vault access can read, search, and follow links until it has what it needs. The depth of context is not bounded by the initial prompt - it is bounded by what exists in the vault and how many tool calls the agent is willing to make.
Setup
Configuration is minimal. The plugin reads the vault path from the OBSIDIAN_VAULT_PATH environment variable, defaulting to ~/Documents/Obsidian Vault if unset. Set the variable in ~/.hermes/.env:
OBSIDIAN_VAULT_PATH=/path/to/your/vault
Once enabled, the agent gets access to the full vault filesystem. No separate database. No index rebuild. The markdown files are the source of truth.
Why the post landed
The numbers on Chawla's post tell a story beyond the engagement counts:
- 1,908 bookmarks - people are saving this as a reference
- 44 replies - active discussion, not passive likes
- 8 quote tweets - others are amplifying and adding their own takes
The post also quote-tweeted a Hermes Agent Masterclass by Akshay Pachaar that had pulled 6,162 likes and 22,487 bookmarks three weeks earlier. The Obsidian integration post landed in an audience already primed to care about Hermes workflows.
Chawla's framing - "a Karpathy-style second brain, but one that talks back" - is specific enough to be useful. Andrej Karpathy's second brain concept describes a personal knowledge base built from markdown files, grep, and linked notes. The Hermes integration takes that idea and adds an active reasoning layer on top of it. The vault does not just store what you know. It becomes context the agent can act on.
What changes
For existing Obsidian users, the integration means their vault is no longer a read-only archive. It becomes a working dataset. The agent can search across years of notes to find a specific fact, follow link chains to build context, and write new observations back into the graph.
For Hermes users who do not currently use Obsidian, the integration lowers the barrier to building a persistent knowledge base. The agent can create notes, link them together, and maintain the graph on its own. The vault grows with use, and the agent gets more useful as the knowledge graph expands.
The filesystem-first approach also means the notes remain fully portable. Any text editor can open them. Any tool that reads markdown can consume them. The agent is an addition to the workflow, not a replacement for the existing tools.
[^1]: Avi Chawla. "The top Hermes integrations to give your agent superpowers." X. June 5, 2026.
[^2]: Akshay Pachaar. "Hermes Agent Masterclass." X. May 13, 2026.