v0.2 · Developer Preview

The versioned filesystem
for AI agents.

Paths, permissions, and history. The primitives every agent already understands, with the governance enterprises actually need.

$cargo install markdownfs
Star on GitHub···Read Docs

grep, across your workspace

Full-text search across every markdown file in the workspace. Find what any agent wrote last week in microseconds.

grep -r 'auth migration' /agents/

commit, revert, inspect

Git-style version control built in. Content-addressable storage deduplicates automatically. Revert any change instantly.

revert --to abc123

chmod, for agents

Unix-style permissions with users, groups, and ACLs. Control which agent can read, write, or own which files.

chmod 640 /agents/billing/

Why a filesystem

When agents outnumber people 10x or 100x, you need infrastructure that proves what they did and stops them from doing what they shouldn't. The questions enterprises ask sound like filesystem questions: who owns this, who changed it, when, and can I roll it back.

MarkdownFS gives every agent a versioned filesystem from day one. Paths, Unix permissions, content-addressable commits, microsecond revert. The primitives every developer already understands, applied to the workloads agents actually run.

“AI agents are going to be the biggest users of software in the future.”

— Aaron Levie, CEO of Box

1.0Unlocks

Build powerful agent systems with just a few lines of code.

Provenance

Every change has an owner

a3f1e8review billing draftalicenow
9c0b22draft Q4 planclaude → alice2m
44a7d1ingest ticketsclaude-bot12m

The commit log is your audit trail. Every write carries an author, timestamp, and content hash. Answer 'who changed what, and when' without leaving the workspace.

Performance

Microsecond commit & revert

commit 10k files3ms
revert 5k files4ms
vs native FS125x

Atomic in-memory core with content-addressable persistence. Commit ten thousand files in milliseconds. Revert thousands of files in microseconds.

Recovery

Checkpoints & rollbacks

rollbackv1v2v3v4v5head

Snapshot the entire filesystem at every step. Roll back to any past state in a single API call. Content deduplication makes commits cheap.

Knowledge

Shared agent workspaces

context.md
skills.md
memory.md
playbook.md
lessons.md
patterns.md

One filesystem, many agents. Each gets a home directory, scoped permissions, and access to shared paths. Read and write across agent boundaries with safety.

Identity

Users, agents & delegation

aliceHuman user
user
claude-botStandalone agent
agent
claude → aliceOn behalf of
delegate

Authenticate as a user, as an agent, or as an agent acting on behalf of a user. Unix groups, ACLs, per-token scopes — and a real delegate command.

Interfaces

CLI, HTTP & MCP

CLI
HTTP
MCP
one core

Three first-class surfaces over the same concurrent core. Use the right interface for the right caller, with no glue layer.

What makes it different

001

Three interfaces, one filesystem

CLI/REPL for interactive use. HTTP/REST API for services. MCP server for AI agents. All share the same concurrent core with atomic persistence.

002

Version control, built in

Git-style commits with content-addressable SHA-256 storage. Automatic deduplication — 100 identical files share one blob. Revert to any state in microseconds.

003

Multi-user permissions

Unix-style users, groups, and ACLs from day one. Each agent gets its own identity, home directory, and API token. Root + wheel for admin privilege.

004

Blazing performance

~125x faster than native filesystem. Commit 10,000 files in 3ms. Revert 5,000 files in 4ms. In-memory core with atomic disk persistence.

Built for these workflows

Agent Workspaces

ls /agents/claude/workspace/

Each agent gets its own home directory with its files, its history, and its permission scope. Pick up exactly where it left off.

Incident Response

mkdir /incidents/2024-03-15/

Structured investigation logs, timeline artifacts, and post-mortems — all versioned and searchable.

Research Notebooks

grep -r 'hypothesis' /research/

Multi-agent research with isolated workspaces. Each agent writes findings, humans review diffs.

Code Review Artifacts

cat /reviews/pr-2847.md

Store review notes, security findings, and approval records as durable, auditable markdown.

Knowledge Bases

tree /docs/onboarding/

Team knowledge that agents can read and update. Version-controlled docs with permission boundaries.

Multi-Agent Collaboration

su agent-billing

Each agent has its own user, home directory, and permissions. Share files safely across agent boundaries.

How it works

Spec Sheet

LanguageRust
Async runtimeTokio
HTTP frameworkAxum
ConcurrencyRwLock (multi-reader)
StorageIn-memory + atomic disk
HashingSHA-256
Tests215 across 5 suites
Lines of code~10,000
Layer 1Surface
  • ·CLI / REPL — interactive terminal
  • ·HTTP / REST API — web apps & services
  • ·MCP Server — AI agent integration
  • ·Remote CLI (mdfs) — thin client over HTTP
Layer 2Core
  • ·MarkdownDb — thread-safe concurrent core
  • ·VirtualFs — inode-based filesystem
  • ·Content-addressable blob store (SHA-256)
  • ·Atomic persistence (bincode → .vfs/state.bin)
Layer 3Middleware
  • ·Auth registry — users, groups, tokens
  • ·Permission engine — Unix ACLs + delegation
  • ·Session context — per-request identity
  • ·Search index — grep + find + tree
Layer 4Version Control
  • ·Commit objects — full tree snapshots
  • ·Content deduplication — automatic
  • ·Revert engine — instant rollback
  • ·Commit log — history, author, timestamp
Early Access

Get early managed access.

MarkdownFS is available today. If you want managed hosting, priority support, or enterprise features, tell us about your team and we'll get you set up.

Or try it right now

Install with a single command. Full source on GitHub. Star it, fork it, contribute.

$cargo install markdownfs
View on GitHub →

No spam. We'll only reach out about MarkdownFS.

mount / agents

Give every agent a versioned filesystem with owners, paths, and history. One command to install.

$cargo install markdownfs
Available Now