Do you find yourself re-explaining last week’s context every time you talk to an AI? Ars Contexta is an open-source Claude Code plugin that tackles exactly that “every session starts blank” problem. Its approach is unusual: instead of handing you a ready-made template, it spends about 20 minutes interviewing you about your work and how you think, then derives a folder structure, note templates, processing pipeline, and automation hooks tailored to your answers. Underneath sits a body of 249 interconnected research claims drawn from Zettelkasten, Cornell Note-Taking, Evergreen Notes, PARA, cognitive science, and network theory — so every piece of the generated system can be traced back to the research that justifies it. It is MIT licensed and was released in February 2026.
Key Features
- Derives a knowledge system from conversation: Rather than distributing a generic template, it runs a six-phase process — Detection, Understanding, Derivation, Proposal, Generation, Validation — to assemble folders, skills, and documentation fitted to your domain. Every design choice links back to one of the 249 research claims, and
/arscontexta:askwill answer questions such as “Why does my system use atomic notes?” with that grounding attached - A three-space architecture: self / notes / ops:
self/holds the agent’s identity, methodology, and goals;notes/is the knowledge graph itself;ops/carries operational state such as queues and sessions. The folder names adapt to your domain —notes/might becomereflections/,claims/, ordecisions/— but the three-way separation is invariant - A “6R” pipeline extending Cornell’s 5 Rs: Record, Reduce (extract insights), Reflect (find connections and update navigation maps), Reweave (fold new context back into older notes), Verify (quality checks), and Rethink (challenge the system’s own assumptions). Each phase spawns a fresh subagent so no single phase drowns in accumulated context
- Hooks that keep the structure intact: Four automated hooks handle injecting workspace context at session start, enforcing schemas on writes, committing to Git, and persisting session state when you stop — so the structure does not quietly drift
- The output is plain Markdown: What you get is a vault of wiki-linked Markdown notes plus multi-level MOCs (Maps of Content). Nothing is locked into a proprietary app, and you can open the same files in Obsidian or any editor you already use
- Semantic search is optional: Adding
qmdenables concept matching across differing vocabularies, but it is not required — the system works fully with ripgrep plus MOC traversal
Pricing
| Plan | Price | Highlights |
|---|---|---|
| Open source (MIT license) | $0 | All features available; published on GitHub, with commercial use and modification permitted |
Pricing is current as of August 2026. Please check the official site for the latest information.
The plugin itself is free, but running it requires Claude Code (v1.0.33 or newer), which is billed separately. Initial setup takes roughly 20 minutes of conversation and is described as token-intensive.
Pros & Cons
✅ Pros
- Designed around how you actually work rather than a template you have to conform to, which helps if past note-taking methods never stuck
- Can explain why a given structure exists, tracing it back to research, so you can keep using it with confidence
- Output is ordinary Markdown, easy to carry elsewhere and not tied to any single vendor
- Hooks maintain the structure automatically, so you spend less time on manual tidying
- MIT-licensed open source, so you can read the internals and adapt them
⚠️ Cons
- Assumes Claude Code, which raises the barrier for anyone who does not work in a terminal
- Requires installing
treeandripgrep, adding a small setup burden - Initial setup costs about 20 minutes of conversation and a fair amount of token usage
- Multi-agent processing is explicitly still in progress, so features may change
- No GUI — it is designed to be operated through commands and files
Comparison with Similar Services
| Criteria | Ars Contexta | Basic Memory | mem0 | CLAUDE.md (built into Claude Code) |
|---|---|---|---|---|
| Form factor | Claude Code plugin | MCP server | Memory-layer library / API | Built-in Claude Code feature |
| Storage format | Markdown knowledge graph | Markdown | Primarily a vector store | Mainly a single Markdown file |
| Structure creation | Derived automatically from conversation | Designed by the user | Designed by the host app | Written by hand |
| Processing pipeline | 6R (six stages) | None | None | None |
| Price | Free (MIT) | Free (open source) | Open source plus a cloud offering | Free (ships with Claude Code) |
Who Is It For
- People who use Claude Code daily and are tired of re-explaining the same context each session
- Anyone who tried Zettelkasten or PARA but gave up because the templates did not fit their work
- Those doing research, writing, or design work where accumulated context compounds over time
- People who would rather keep their notes as local Markdown than hand them to a vendor
- Readers who want to understand the reasoning behind a system before committing to it
Summary
Ars Contexta implements the idea of “giving your AI a memory” not as a generic template but as a derivation performed for each individual. The skeleton — the self / notes / ops separation and the 6R pipeline — stays fixed, while the contents are assembled from your own answers, which gives a way in even for people whose earlier note-taking attempts fizzled out. The trade-off is that it presumes a terminal and Claude Code, and asks for an up-front investment of about 20 minutes of conversation plus some token spend. Since it is free under the MIT license, a reasonable approach is to read the GitHub documentation first, get a feel for the folder structure it generates, and decide from there.