obsidian-wiki

A Python framework that lets an AI agent compile and maintain an Obsidian wiki for you. You wire it to whatever content source you have (Slack, transcripts, meeting notes, emails); it distills that into a structured Obsidian vault with page-level cross-links. Your vault becomes a textbook of what your organization is learning.

good for

Python-capable teams that already use Obsidian
Founders who want their meeting transcripts / Slack threads auto-compiled into a readable wiki
Anyone implementing the Karpathy pattern on an existing Obsidian vault

Skip this if: You don't have Python on the team, or you don't already use Obsidian.

What you'll do

obsidian-wiki reads from content sources you configure, runs an LLM-based compile step, and writes structured Markdown into your Obsidian vault. Budget 25 minutes including configuring your first source.

Before you start

Python 3.10+
An existing Obsidian vault (or create a new empty one)
An OpenAI or Anthropic API key (for the compile step)

Step-by-step install

01
1. Install the package
Use pip. We recommend a virtualenv.
```
pip install obsidian-wiki
```
02
2. Point it at your vault
Create a config file telling obsidian-wiki where your vault is and which subfolder should receive generated pages (keep these separate from your hand-written notes).
```
# ~/.obsidian-wiki/config.yaml
vault_path: /path/to/your/vault
output_folder: compiled/
api_key: sk-...
```
03
3. Configure a content source
obsidian-wiki ships adapters for common sources: transcripts (Granola, Fathom, Otter), Slack exports, email archives. Pick one and wire it in the config.
Tip: Start with one source. Don't try to ingest everything on day one — you'll drown in LLM costs and get a messy first wiki.
04
4. Run a compile
Run the compile command. It reads from your source, extracts concepts, and writes/updates pages in your vault.
```
obsidian-wiki compile
```
05
5. Open Obsidian and browse
Open your vault. Browse the output folder. You should see structured pages about the people, topics, and decisions from your source.
06
6. Connect your agent to the vault
Install obsidian-mcp-tools so your Claude Code or Cursor agent can read (and talk to) the wiki you just built.

Your first 10 minutes

01Feed it one week of meeting transcripts. See what it compiles.
02Hand-edit a generated page. Run compile again and confirm it respects your edits.
03Add a second source. Watch how it cross-links concepts.
04Git-version the vault. You now have an audit trail of what your org learned, per commit.
05Add Cognition CLO so the wiki's concepts get retention tracking per team member.

Troubleshooting

Compile step is very expensive.

Configure it to use a cheaper model for extraction (e.g. gpt-4o-mini or haiku). The compiler's docs have model-swap examples.

Generated pages are low quality.

Pass more structured source data. Raw Slack is worst; formal meeting notes are best. Clean input, clean output.

once obsidian-wiki is installed

Layer CLO on top and start modeling what your team is about to forget.

obsidian-wiki holds the knowledge. Cognition CLO models retention per employee per concept using a Weibull forgetting curve — so you see decay before it becomes a missed SOP or a failed audit.

Build my full stack in 5 steps →Back to registry