// Self-hosted sync infrastructure

Your team and your AI agents work from one shared brain.

Raven Sync keeps your company knowledge base in sync as plain files on every machine. Concurrent edits merge automatically, each member sees only the folders you grant, and every file stays on disk.

automatic merge · per-folder access control · files stay on disk

Sync your vault See how it works

the CEO's machine: company-kb/

company-kb/
├ strategy/     plans · decisions · processes
├ marketing/
├ engineering/
├ clients/      one folder per client
├ people/
└ harness/      claude · codex: agents work here

My whole company runs from our markdown vault. But how do I share it with my team?

1 Command to join

3-way Automatic merge

0 Telemetry

100% Files on your disk

// The setup

Set it up in one conversation.

sarah needs marketing/, mike needs engineering/, and the CEO needs to see everything. Ask your AI assistant: it runs the real Raven Sync CLI.

admin machine: one conversation

● you
  set up our company vault. marketing and engineering each get
  their folder to edit, strategy/ is read-only for both.
  then add sarah and mike.

● assistant
  ✳ loading skill: ravensync
  $ ravensync vault create acme
  $ ravensync team create acme marketing
  $ ravensync grant add --actor ceo acme marketing marketing/ write
  $ ravensync grant add --actor ceo acme marketing strategy/ read
  … engineering set up the same; sarah and mike added, tokens issued
  ✓ vault ready: 2 teams, 2 members, per-folder rights in place

  Send sarah and mike their enroll commands. One paste on their
  machine and their folders appear on disk, synced at every login.

Claude Code, Codex, or any agent works: it is a real CLI. Members join with one command that authenticates first and only then writes to disk; a wrong token fails immediately and leaves nothing behind.

// How it works

Work with synced local files.

The fastest way for humans and AI agents to work together.

The admin sets up the vault

Create the vault, teams, and per-folder grants on one server. No cluster, no external services, nothing else to run.

A member runs one command

enroll writes the config and token, registers login autostart, and runs the first sync. The same command on macOS, Linux, and Windows.

Files flow to disk

Each member receives only what they are granted, filtered at distribution. Concurrent edits merge automatically. It is not a web app: it is your disk.

// What it does

Infrastructure, not a knowledge app.

A sync engine with the guarantees a company brain needs.

Automatic 3-way merge

Concurrent edits to the same markdown, YAML, or JSON merge key-by-key; code and text get a line-level merge. Both edits survive, and a genuine overlap never creates a file (conflict copy).md sibling: it flags the file and keeps every version in history.

Per-folder access control

Team-based grants enforced server-side on every read path: pull, version history, and the conflicts listing. A revoked grant cuts off old versions too, not just future pulls.

Files on disk, agent-native

Every editor and every AI agent works with the files natively. No editor, no UI, no semantic layer to work around.

One-command onboarding

enroll handles config, token, login autostart, and first sync on macOS, Linux, and Windows. The daemon restarts on crash and starts at every login.

Nothing rewritten in your files

No injected IDs, markers, or metadata. All bookkeeping lives out-of-band. A file the engine never merged is byte-identical on every machine, so Obsidian and OKF vaults stay intact.

Self-hosted, no telemetry

A self-contained service on your own infrastructure: no external database, no runtime dependencies. It never calls home: no analytics, no cookies, no phone-home.

// Skills and agents, company-wide

Share and upgrade your agents and skills in real time.

Your AI agents are only as good as the skills and instructions they run. Raven Sync distributes those files like code, with control.

harness/ on every member's machine

harness/
├ claude/
│   ├ CLAUDE.md    company-wide instructions
│   ├ skills/      write-report · onboard-client
│   └ agents/      researcher · critic
└ codex/         same brain, different harness

One update, every machine

Improve a prompt, a skill, or an agent config once. Every member's machine has the new version in seconds, no copy-paste, no stale wiki.

Any harness, same brain

Claude Code, Codex, or whatever comes next: agents on every machine read the same instructions, the same skills, and the same knowledge base.

Governed like code

A folder carrying executable skills syncs read-only for members by default. Only maintainers you name can change it, every write is audit-logged, and append-only history rolls a bad update back.

Propose, don't push

Members without write access still contribute: the starter vault ships a proposal flow, so improvements get filed, reviewed by a maintainer, and applied with attribution.

// Why not X

Where the alternatives stop.

No product ships all three of concurrent-edit merge, per-folder access control, and local files on disk. Each alternative solves at most two.

Capability	Syncthing	Dropbox / Drive	Bare Git	Notion-class	Raven Sync
Files on disk, agent-native	✓	~	✓	✗	✓
Concurrent edits merge, both survive	✗	✗	~	✓	✓
Per-folder team access control	✗	✗	✗	✓	✓
Non-technical members welcome	~	✓	✗	✓	✓
Self-hosted on your infrastructure	✓	✗	✓	✗	✓

✓ yes · ~ partial or manual · ✗ no

// Security

What it will never do.

Your company brain is the most sensitive thing you own. These guarantees are enforced in the engine, not promised in a policy.

SECURITY.md

## leak nothing

An unauthorized request gets an answer indistinguishable from "does not exist", so vault names cannot even be probed. Access control runs server-side on every read path: sync pull, version history, conflict listings.

## revoke means revoke

A revoked grant cuts off old versions and conflict entries too, not just future syncs.

## code is governed

Syncing scripts means distributing code to your members' machines. Folders carrying executable files therefore sync read-only by default, and granting write access there is an explicit, audit-logged decision.

## nothing vanishes, nothing calls home

History is append-only with an audit trail; every change has an author. The binaries carry no telemetry.

## exit stays open

Plain files on disk plus a one-way git mirror. You leave with everything, any time, in under an hour.

Run it inside your own perimeter, or on a dedicated EU instance with a DPA. The guarantees hold either way.

// Works with what you already use

Because it is just files.

Raven Sync adds no format and no lock-in. Whatever reads a folder of files already works.

Obsidian
Claude Code
Codex
VS Code
Any editor
Any AI agent

Any file type syncs. This is how each kind is handled when two people edit at once:

Markdown, YAML, JSON

First-class knowledge formats: concurrent edits merge key-by-key, YAML frontmatter included, and both edits survive. Untouched parts keep their exact formatting, including key order, quoting, and comments.

Code, scripts, any plain text

Line-level merge, the way git resolves parallel work. Prompts, configs, shell scripts, source files.

Images, PDFs, binaries

Sync unchanged. When two machines race, the newest version wins and every earlier version stays in history.

// Get started

Deploy it with us or on your servers.

Two ways to run Raven Sync. Deploy it on your own infrastructure with our guidance, or try a hosted instance first; we reply personally.

Hosted instance

A dedicated instance on EU infrastructure with backups and an audit log, so you get the product without running the server yourself. Free 14-day trial: we provision your instance and send you everything you need to connect, usually within one business day.

Self-hosted, with guidance

We deploy Raven Sync on your own infrastructure and help you design teams, grants, and member enrollment. Your data never leaves your perimeter.

Name

Company

People sharing the vault

Deployment

What are you looking to sync?

// Questions

Before you ask.

The objections we hear most. If yours is missing, ask in the form above; we reply personally.

Does it work with Obsidian and our existing vault?

Yes. Raven Sync adds no format and rewrites nothing: no injected IDs, markers, or frontmatter. A file the engine never merged stays byte-identical on every machine, so an existing Obsidian or OKF vault syncs as-is.

What happens when two people edit the same file?

Concurrent edits merge automatically: key-by-key for markdown, YAML, and JSON, line-level for code and plain text. Both edits survive, and a genuine overlap never produces a conflict-copy sibling. The file is flagged for review and every version stays in history.

Where does our data live?

On your machines and your server. Self-host Raven Sync inside your own perimeter, or use a dedicated EU instance with a DPA. Either way the binaries carry no telemetry and never call home.

How do we migrate off it?

Your files are already plain files on disk, and the built-in git mirror keeps a one-way backup. You leave with everything, any time, in under an hour.

Your vault, on every machine.

Hosted EU instance with a 14-day trial, or self-hosted inside your perimeter. Either way, your team and your agents share one brain.

Sync your vault See how it works