Rootmoss

Your AI keeps forgetting you. Your files don't.

Your AI's memory, as a folder on your computer.

ChatGPT, Grok and Claude all remember what you've told them — but that memory lives inside their app, on their servers. Rootmoss is the same idea: a folder of plain files on your own disk, read by the AI you already use. It stays yours, and nothing gets deleted — notes get marked, not lost.

An AGENTS.md template for a personal AI memory — a folder your agent reads at session start, so it remembers you across chats, models and tools. See the landing page.

Every chat starts from zero: you re-explain who you are, what you decided, what matters to you. Rootmoss gives any AI a memory that doesn't reset — a folder of plain files it reads at session start. Your memories stay on your disk, in files you can copy and back up — they never get locked inside a chat vendor's memory. Change the model, the tool, even the company behind it — your history comes with you.

• Better conversations from the first message — your agent starts with your profile, your open topics, what you learned last time.
• Truly yours — plain files, no Rootmoss account, no subscription, no lock-in: your data never lives only inside a vendor's product.
• It compounds — chats stay disposable; what you save keeps paying off as your tree grows.

Works with OpenCode — a free, ChatGPT-style app you install — in terminal, desktop app or web, or with anything else that reads AGENTS.md. Open the folder like any folder: Finder is enough. Obsidian or Zed is optional, if you want a nicer view.

Status: experimental / early. The name and logo are not covered by the license.

What you get

Piece	What it does for you
AGENTS.md	The contract your agent reads at session start
opencode.json	The permission floor — what it may do without
	asking you (edit: ask, bash: ask)
FIRST-TIME-HERE.md	Full tour: concepts, FAQ, first steps
USER.md	Your thin profile — the agent's starting point
memory/<6 types>/	Typed folders: facts, preferences, decisions,
	assumptions, lessons, context
inbox/	Staging for chats that fit no topic yet
topics/	One folder per life subject, each topic.md
skills/	Portable procedures your agent follows
topics/plants/	A fictional demo topic showing the pattern

Prerequisites

• OpenCode — desktop app, terminal or web; about 5 minutes to install
• git — or skip it and use Download ZIP (updates get harder without git)
• Optional: Obsidian or Zed — Finder is enough

First run takes about 15 minutes.

Quickstart

Not technical? Only steps 1–3 need anyone technical, and they happen once. Step 4 is yours, and step 4 is just pasting a message. Everything after that is conversation.

1. Get the files. Two ways:

   • No GitHub account: download the ZIP and unzip it — no login at all. Go straight to step 2.
   • Using git: click Use this template → create your own private repo, then clone it. Your host still stores the files, so keep raw sensitive documents (bank exports, scans, IDs) out of any cloud repo. Your copy is a snapshot: it only changes when you pull.

   Download the ZIP · or git clone <your-repo-url> rootmoss && cd rootmoss

2. Install OpenCode — pick whichever suits you:

   • Desktop app (no terminal): download from opencode.ai and install.

   • Terminal:

     brew install anomalyco/tap/opencode-v2   # macOS (recommended)
     npm install -g @opencode/cli             # any platform
     curl -fsSL https://opencode.ai/v2/install | bash   # Linux/macOS
     

   • Web: OpenCode also runs as a web app — needs a one-time setup by someone technical; after that, daily use is just a browser tab.

3. Start it from inside the folder, and pick a model:

   • Desktop app: open Projects → Add project and choose your rootmoss folder. OpenCode calls it a project — it is just this folder. (In a terminal: run opencode from inside it.)
   • Model: click the model name under the prompt box, then Connect an integration — sign in with ChatGPT if you already pay for it, or pick any provider you have. Terminal: /connect, then /models.

4. Paste your first message:

   Read AGENTS.md and FIRST-TIME-HERE.md. Then interview me one
   question at a time to fill in USER.md, and propose my first topic
   based on what I tell you.
   

Then work through FIRST-TIME-HERE.md — folders, memory types, and the rules your agent follows.

Who this is not for

• You want a polished app with accounts and sync — this is a folder.
• You never want to set anything up — daily use can be terminal-free (desktop app or web), but the one-time setup still takes about 15 minutes; have someone technical do it once.

Storage is your choice

Rootmoss is a folder — it doesn't host, sync, encrypt or back up anything. Where your copy lives (local disk, GitHub, a NAS, self-hosted git), what you back it up to and what you encrypt is up to you. The docs recommend patterns; no host or tool can guarantee them. If your notes matter, keep more than one copy: the classic rule is 3-2-1 — three copies, two kinds of media, one offsite. A good plain-language explanation: Backblaze on the 3-2-1 rule.

The backup part is on you — same for the AI provider you connect: your account, your terms. One thing always travels: when your agent reads a note, that text goes to the model you connected, exactly as it would in any AI chat. Your folder stays put; the words it reads travel.

Support

Experimental. "Where did I get stuck?" issues are the most welcome contribution; responses are best-effort, not guaranteed.

Updates

Releases are announced on the repo's Releases page. To get an email per release: Watch the repo → Custom → Releases → Notify me when releases are published. Or check CHANGELOG.md before pulling (FIRST-TIME-HERE.md §8).

License

MIT — code and prose. The Rootmoss name and logo are not covered by the license; please don't publish confusingly similar derivatives under this name.