I · brain
Start here. Plain markdown files your assistant reads and updates when instructed. You can open and edit them yourself.
A home for your AI assistant’s memory.
Keep your projects, preferences, and decisions in files you control. Start with one project and configure your assistant to carry useful context between sessions.
Free template. Download the files, then set it up with your coding assistant.
The empty cathedral. You bring the life.
Build a system that is actually yours.
Your saved context lives in plain files you control. Lararium gives your coding assistant a structure for reading and updating them. Start with one project and add more as it becomes useful. Your chosen AI provider still processes the context you send it.
Start with the brain. Add what you need.
II · soul
Optional. Define your assistant’s voice, working preferences, and boundaries in a file you can revise.
III · index
Optional. Schemas and guidance for searching saved context. You supply and run the database and search service.
IV · agents
Optional. Define specialists for tasks such as building, reviewing, and research, using your assistant’s supported tools.
V · hooks
Optional. Configure supported session hooks to save a short handoff and load it next time. Verify them in your own setup.
VI · skills
Optional. Reusable instructions for recurring tasks, saved as files your assistant can consult.
The install is a conversation
Get the files. Open your assistant. Make it yours.
The command downloads the template. Then open the folder in Claude Code and ask it to run the install interview. Start with the brain: a few real notes and a map of what matters. The other layers are optional.
You need Node.js 18 or newer, tar, and a coding assistant with local file access. Claude Code is the reference setup and needs its own subscription or API access. Git is optional. Lararium itself is free.
The lab
Inspect unfamiliar code in a disposable lab.
The optional lab requires Docker. It clones the repository with network access, then runs static analysis offline in a disposable container with no host folders mounted and reduced privileges. Use lab --analyze to inspect code without executing the repository’s own scripts. A container reduces exposure; it is not a guarantee that code is safe.
Want to inspect Lararium first? The scaffolder uses Node’s standard library with no dependencies. It downloads and unpacks the template without running the template’s code. Read the scaffolder, or clone the repository yourself.
Agent-native
Hand the whole thing to your agent.
The methodology fits in one reference file: what Lararium is, how setup works, and why each layer exists. Share it with your assistant to plan a setup that fits your work. Installing and maintaining it requires a coding assistant with access to your local files.
A reference for planning. Download the template when you are ready to build.
It gets sharper over time
Useful context gets more useful when you return to it.
Use the included guidance to build a routine that connects saved notes to current work. You configure this optional layer with your assistant.
Illustrative example
A note you wrote three weeks ago, "consistency vs. availability," just connected to the doc open in front of you.
Stay in the loop
New capabilities ship often.
Get occasional release notes and practical setup tips. The optional update hook can also check for new versions in your configured stack. Unsubscribe any time.
Want this built and run for you, not by you? That is what Elorati does.