For Claude Code + Obsidian

A strategy and execution layer for solo founders and small teams.

FonteOS gives Claude Code persistent memory, project awareness, and session continuity — through a vault structure you clone, not a service you subscribe to.

No server. No account. No build step.  —  Clone, open, /start.

Claude forgets everything
between sessions.

You're re-explaining your world every time you open a new conversation. That's not a workflow — that's overhead.

01

"You re-explain your context every session."

Your mission, your projects, your priorities, your decisions. Every conversation starts cold. You spend the first 10 minutes rebuilding what Claude should already know.

02

"Your projects bleed into each other."

No isolation means Claude mixes concerns between ventures, clients, or products. A question about one project pulls in noise from another. Signal degrades.

03

"You burn tokens re-loading what Claude should already know."

Dumping your notes into every session is expensive and slow. You're paying to re-teach, not to work. There's no structure — just bulk paste and hope.

Clone the repo.
Run /start. Pick up where you left off.

FonteOS is a vault structure and a set of behavioral rules — written in plain markdown. Claude loads it automatically every session.

01

Clone the repo

It's just markdown files and a CLAUDE.md. No install, no dependencies, no configuration wizard. Open the folder as an Obsidian vault or any markdown editor. Done.

git clone → open → done
02

Run /start

Claude reads your world: mission, active projects, last session state. It cross-references your task manager, surfaces what's relevant, scopes to one project. The first run walks you through setup. Every run after that is instant.

/start → context loaded → work begins
03

Work → /end → next session picks up cold

At the end of each session, /end saves your state: decisions made, work done, what's next. The next session reads that state and starts where you left off. No recap needed.

/end → state saved → cold start eliminated

Six layers. One source of truth.

Every piece of context has a home. The structure tells Claude exactly what to load — and what to leave alone.

Source core.md Mission · priorities · registry
Springs projects/ One index per project
Streams stream/*.md Active workstreams
Intelligence intelligence/ Research · reference · confidence-rated
Agents agents/ AI role definitions
Inbox inbox/ Loose ideas · triaged weekly
~5k
tokens of context per session. 90% of your window stays free for work. Claude loads what's relevant — not your entire life.
Source (core.md) loaded on /start.
Spring index loaded when you pick a project.
Streams loaded as needed.
Intelligence loaded on demand.

External data (tasks, research, docs)
stays external until you need it.

Built for how solo operators
actually work.

No feature bloat. Everything here exists because running multiple projects with Claude, without it, is painful.

/start

Session continuity

Last session state is saved and reloaded automatically. Cold starts eliminated. Pick up exactly where you left off, every time.

[ ]

Spring isolation

Each project lives in its own spring with its own index and streams. Claude scopes to one project per session. No bleed-over, no mixed context.

Graduated knowledge

Stream insights that recur 3+ times get promoted to Intelligence files — permanent, confidence-rated reference knowledge that outlives any session.

Write permissions

Claude writes freely to streams and inbox. Everything permanent — source, intelligence, spring indexes — requires human approval. You control what crystallizes.

Token-efficient context

The hierarchy means Claude loads ~5k tokens of context, not your entire vault. Targeted loading. 90% of the window stays free for actual work.

Optional MCP integrations

Works standalone. Gets more powerful with Asana or Linear for task sync, Perplexity for deep research, Gemini for document analysis, Figma for design files.

No hype. No dependencies.
No surprises.

×

Not a chatbot

FonteOS is a working context for Claude Code — a structured environment, not a conversation interface. It assumes you're already in the terminal.

×

Not a RAG pipeline

No vectors. No embeddings. No similarity search. Just markdown files loaded strategically through a hierarchy Claude knows how to navigate.

×

Not a SaaS

No server, no account, no subscription, no API key to manage. It lives in a folder on your machine. You own it entirely.

×

Not an installer

It's a file structure. Clone it and open it. No npm install, no setup.sh, no wizard. If you can read markdown, you can understand every part of it.

Four commands.
You're running.

First run walks you through setup. Every run after that loads your world in seconds.

zsh
$ git clone https://github.com/ShrimpPasta/fonteos.git my-vault
Cloning into 'my-vault'...
$ cd my-vault
$ claude
Claude Code ready.
> /start
Reading core.md... Loading session state... Checking integrations...
Context loaded. What are we working on?