Mimir — a living wiki in Obsidian, powered by Hermes
A living wiki in Obsidian, powered by Hermes to connect, organise and recall everything you know.
Mimir is a second brain for Obsidian: an agent, driven by Hermes, that collects, links and organises your notes into a living wiki — to recall everything you know.
The idea
A second brain that tends itself. You hand it whatever you want to keep — an article, a document, a note, an idea caught in passing — and rather than letting it pile up in a heap, Mimir digests it: it links it, shapes it, files it away. A living body of knowledge that grows and weaves itself over time.
The goal isn't to hoard, but to understand: turning a stream of scattered sources into clear, navigable, connected knowledge.
The raw material
Everything you've gathered, kept exactly as it was, intact. A faithful trace of what you came across — never altered.
The built knowledge
Re-read, rephrased, structured, linked. What you actually took away and understood, filed by notion.
Mimir bridges the two, continuously — without ever copying a raw source into the wiki: it understands it first, then rephrases it.
How it works
Under the hood, Mimir is a suite of 8 skills (agentskills.io) orchestrated by the Hermes agent. A source goes in, a connected wiki comes out:
Source (PDF · EPUB · URL)
│
├─ wiki-extract → stores the raw source, faithful and immutable
├─ wiki-ingest → digests it into wiki articles, one notion per page
├─ wiki-batch-produce → same, in deterministic batch (web only)
├─ wiki-reading-grid → derives an ordered reading grid from it
├─ wiki-index → rebuilds the INDEX and checks the links
└─ wiki-sync → syncs everything to the Obsidian vault
A seventh skill, wiki-init, sets the stage on first use: it creates the second brain's folder structure and bootstraps the sync.
An eighth skill, wiki-research, lives in a different lane: for a deep-dive on a text subject, the agent picks web sources, has them clipped automatically (it reuses wiki-extract), then writes a synthesis report in reading-grids/<subject>.md. Source picking and writing are semantic; the clip and the scaffold stay deterministic.
Installation
Prerequisite: Hermes Agent installed.
hermes profile install https://github.com/vivian-maes/mimir.git
hermes update # seeds the base skills: without it, the profile only has the wiki-*
On first launch, on a Hermes that isn't configured yet, the “hermes setup” assistant configures the provider, the terminal and — optionally — messaging:
hermes -p mimir chat
Finally, drop wiki.config.json into ~/.hermes/profiles/mimir/ (auto-discovered), then check with hermes profile info mimir.
Updating (profile already installed): hermes profile update mimir — the command takes the profile name, not the git URL.
First use
On an empty vault, ask Mimir to “initialise the wiki” (the wiki-init skill): it creates the folder structure (_inbox/, raw/, wiki/, reading-grids/), a quick-start guide, a starter INDEX.md, then bootstraps the first sync. After that, dropping a source into _inbox/ kicks off the pipeline.
A note on language
Mimir was born in French. I'm a French speaker (Belgian, mind you — not French, let's not start anything), and it began as a personal project I'm sharing in the language I thought it up in.
This English version is a fresh translation — if a sentence reads a little sideways, that's on me. Full multilingual support is for later: somewhere between “soon” and “when the fries are done”. 🍟
It's young, it's being debugged, it'll move — and any feedback is welcome: a bug you hit, an idea, a use case nobody thought of. Open an issue or share your experience.