FonteOS

A strategy and execution layer for solo founders and small teams using Claude Code.

FonteOS gives Claude Code persistent memory, project awareness, and session continuity. It's a vault structure + behavioral rules — no server, no database, no build step. Clone it, open it, /start.

---

What it solves

• You re-explain your context every Claude session
• You lose decisions, research, and progress between conversations
• You burn tokens re-loading context that should already be there
• Your projects bleed into each other with no isolation
• You have no structure connecting strategy (what to build) to execution (what to do next)

How it works

Two control files:

• CLAUDE.md — behavioral rules Claude reads automatically. Session protocol, write permissions, standing rules.
• source/core.md — your operating context. Mission, priorities, project registry. Claude reads it, knows your world.

Six-layer vault:

• Source — mission, priorities, registries (updated weekly)
• Springs — projects/ventures, each with their own index
• Streams — active workstreams within a spring (where daily work happens)
• Intelligence — research-backed reference knowledge (graduated from streams)
• Agents — AI agent role definitions
• Inbox — loose ideas, triaged weekly

Session lifecycle:

/start → Claude loads your context, suggests work, scopes to a project
  ↓
work → Claude reads/writes within the scoped project, fetches external data via MCP
  ↓
/end → summarize, upstream insights to permanent files, save state for next session

Token-efficient by design. The vault structure means Claude loads ~5k tokens of context, not your entire life. MCP integrations fetch live data from your task manager, docs, and research tools on demand — external data stays external until needed.

---

Quickstart

1. Clone

git clone https://github.com/ShrimpPasta/fonteos.git my-vault
cd my-vault

2. Open in Obsidian

Open the cloned folder as an Obsidian vault. (If you don't use Obsidian, any markdown editor works — you'll just miss the graph view.)

3. Start Claude Code

claude

4. First run

/start

On first run, Claude detects an empty core.md and walks you through setup:

• Who you are and what you're building
• Your top priorities
• Which task manager to connect (Asana, Linear, or none)
• Creates your first project from priority #1

After setup, every future /start loads your context instantly.

---

Vault structure

my-vault/
├── CLAUDE.md                          ← Behavioral rules (auto-loaded)
├── source/
│   ├── core.md                        ← Mission, priorities, registries
│   └── session-state.md               ← Last session pickup
├── springs/
│   └── my-project/
│       ├── my-project_index.md        ← Project overview
│       └── stream/
│           └── .gitkeep
├── intelligence/
│   └── .gitkeep
├── agents/
│   └── .gitkeep
├── inbox/
│   └── .gitkeep
└── .claude/
    └── commands/
        ├── start.md
        ├── end.md
        ├── audit.md
        └── note.md

---

Key concepts

Concept	What it means
Source	core.md — your mission, priorities, and project registry
Spring	A project or venture. Has an index and streams.
Stream	Active work within a spring. Claude writes here freely.
Intelligence	Research-backed reference. Slow-changing, confidence-rated.
Upstream	Codify session output into a permanent file.
Dispose	Close a session without extracting anything.
Graduate	Promote a recurring stream insight to Intelligence.

---

Write permissions

Layer	Claude can write?	Notes
Streams	Yes	Daily work, drafts
Inbox	Yes	Catching ideas
Session state	Yes	Auto-written at /end
Spring indexes	Propose only	Human approves
Intelligence	Propose only	Human approves
Source (core.md)	Propose only	Human approves
CLAUDE.md	Never	Human-only

---

Optional integrations (MCP)

FonteOS works standalone, but gets more powerful with MCP integrations:

Integration	What it does
Asana / Linear	Pull tasks at /start, update at /end, daily cycle automation
Perplexity	Deep research without leaving the terminal
Gemini	YouTube video extraction, document analysis
Google Drive	Read spreadsheets and docs as intelligence sources
Figma	Read design files and extract design tokens

See docs/mcp-integrations.md for setup guides.

---

Commands

Command	What it does
/start	Load context, suggest work, scope to a project
/end	Summarize session, upstream or dispose, save state
/audit	Monthly maintenance — find orphans, stale streams, drift
/note	Create a new note with proper structure and wikilinks

---

What it's NOT

• Not a chatbot. It's a working context for Claude Code.
• Not a RAG pipeline. No vectors, no embeddings. Just markdown files loaded strategically.
• Not a SaaS. No server, no account, no subscription.
• Not an installer. It's a file structure. Clone and go.

---

Documentation

• Getting Started
• Core Concepts
• Session Lifecycle
• Customization
• MCP Integrations
• Voice Profiles
• Advanced: Multi-session Work
• Advanced: Daily Cycle Automation

---

License

MIT

---

Credits

Built by Luke in Lisbon. Born from running multiple ventures with Claude Code and needing a system that doesn't forget.

"Fonte" is Portuguese for spring — the source where streams begin.