Ars Contexta: How I Connected Obsidian with Claude Code and Stopped Losing Context
If you keep notes in Obsidian and work with Claude Code — sooner or later the question arises: how do you make these two worlds talk to each other? How do you get context from notes into the agent, and insights from sessions back into the vault?
I tried several approaches: from manual copy-paste to complex configs. And finally found Ars Contexta — a Claude Code plugin that solves this task more elegantly than anything I've seen.
What Is Ars Contexta
Ars Contexta is a Claude Code plugin that generates a personalized knowledge management system through conversation. Not a template, not a ready-made framework — but a derivative of how specifically you think and work.
The name references historical mnemonic systems: Ars Combinatoria and Ars Memoria. The project philosophy: this is not storage, it's a tool for thinking. And now an LLM can traverse it.
The gist: you describe your domain, answer 2-4 questions, and in ~20 minutes you get a full second brain — folder structure, note templates, processing pipeline, automation hooks, and a 7-page manual.
Architecture: Three Spaces
Every generated system is built on three invariant spaces:
- -
self/ — persistent agent identity, methodology, goals (slow growth: tens of files)
- -
notes/ — the knowledge graph core (steady growth: 10-50 files weekly)
- -
ops/ — operational state, queue management, sessions (fluctuating)
Names adapt to your domain: e.g., "reflections/" instead of "notes/".
Processing Pipeline: The Six Rs
Extending the Cornell Note-Taking method. Each phase runs as a separate subagent with a clean context window:
- -
Record — zero-friction inbox capture
- -
Reduce — extract domain-native insights (/reduce)
- -
Reflect — find connections, update MOCs (/reflect)
- -
Reweave — backward pass updating older notes with new context (/reweave)
- -
Verify — description + schema + graph health checks (/verify)
- -
Rethink — challenge the system's own assumptions (/rethink)
Key detail: each phase spawns a fresh subagent. LLM attention degrades as context fills — splitting into phases keeps every operation in the "smart zone."
My Journey: From VS Code to the kb Alias
I used to manage notes in VS Code — opening the vault as a project, using search, editing markdown files. It worked, but the context between notes and code remained disconnected.
Then I configured Cowork from Claude Code for this task — essentially creating an agentic mode where Claude could read and write notes in the vault. It was better: the agent could see context, suggest connections. But the configuration was fragile, and every new vault needed fresh setup.
Then I found Ars Contexta — and that was the moment everything clicked. The plugin doesn't just give you tools — it generates an entire cognitive architecture tailored to you. Note context and connectivity improved dramatically. Notes stopped being a dump of thoughts and became a working system.
Where to Run Ars Contexta
This is an important practical point I learned the hard way:
Don't run it from your project folder!
- -
Ars Contexta creates notes/, ops/, inbox/ — they pollute your project
- -
Files end up in git (unless you add them to .gitignore)
- -
Knowledge structure mixes with code — defeating the purpose of separation
Best to run from the home directory (~) or directly from your vault folder:
Or create an alias in ~/.zshrc:
Now kb in any terminal opens Claude Code straight in the vault. All /reduce, /reflect, /reweave and other commands will work with the right files.
Tips for Working with Obsidian
Wiki links work out of the box
Ars Contexta generates notes with [[wiki links]]. Obsidian picks them up automatically — the graph becomes visible and clickable.
Graph View as a knowledge map
After a few /reflect and /reweave sessions, your Obsidian Graph View transforms into a real connection map. MOC files (Map of Content) become hubs with thematic clusters radiating from them.
Inbox → Obsidian → /reduce
Quick notes from Obsidian Mobile land in the inbox. Then open Claude Code in the vault, run /reduce — and raw thoughts become structured insights with proper tags and connections.
Session hooks = autopilot
On Claude Code start, the plugin automatically injects the workspace tree and loads agent identity. On exit — saves session state. You don't lose context between sessions.
Brain state export
Everything is plain markdown. No database, no cloud, no vendor lock-in. If Ars Contexta disappears tomorrow — your notes remain readable and connected.
Key Commands
After setup, you get an arsenal of commands:
- -
/arscontexta:setup — conversational onboarding (~20 minutes)
- -
/arscontexta:health — vault diagnostics
- -
/arscontexta:help — contextual guidance
- -
/reduce — extract insights from inbox
- -
/reflect — find connections between notes
- -
/reweave — update old notes with new context
- -
/verify — check graph integrity
- -
/pipeline — run full processing cycle
- -
/graph — visualize structure
- -
/stats — vault statistics
249 Research Claims Under the Hood
One thing that hooked me: every architectural decision is backed by specific research. Not "we decided so" but "here's why."
The methodology/ directory contains 249 interconnected research claims spanning:
- -
Zettelkasten, Cornell Note-Taking, Evergreen Notes
- -
PARA and GTD methodologies
- -
Memory Palace techniques
- -
Cognitive science (extended mind, spreading activation, generation effect)
- -
Network theory (small-world topology, betweenness centrality)
- -
Agent architecture (context windows, session boundaries, multi-agent patterns)
Example: MOC file hierarchy derives from context-switching cost research (Leroy, 2009).
Conclusion: From Chaos to System
My path went like this: VS Code for manual editing → Cowork from Claude Code as an agentic mode → Ars Contexta as a full cognitive architecture. Each step removed barriers between thinking and recording.
If you keep notes in Obsidian and work with Claude Code — give it a try. Setup takes 20 minutes, and you feel the difference from the first /reflect session.
Tip: create a kb alias, launch Claude Code from your vault, and let the agent work with your notes as a living system, not just a folder of files.
Links
Ars Contexta — GitHub
https://github.com/agenticnotetaking/arscontexta
Ars Contexta — project site
https://www.arscontexta.org/
Building your own knowledge system or looking for ways to connect Obsidian with AI?
Get in Touch