Your agent's favorite filesystem.

Markdown filesystem for agents and teams.

Searchable, structured, versioned markdown. Plug in an agent in 30 seconds — or self-host with a single binary.

Get Started Free Self-host

Open Source · BSL-1.1 · Cloud & Self-hosted

KiwiFS mascot relaxing on the beach

Install in seconds

One command. Self-hostable.

$ curl -fsSL https://...kiwifs/main/install.sh | sh

$ kiwifs init --root ./knowledge

$ kiwifs serve --root ./knowledge

Server running at http://localhost:3333

Why we love KiwiFS

And you would too.

One binary. Zero deps.

One binary. Zero deps.

Download, run, done. No runtime, no database, no infra. Ships the web UI, search engine, and git layer inside one Go binary.

$ cat concepts/auth.md

# Authentication

OAuth2 + JWT flow...

$ grep -r "timeout" /kiwi/

config.md:timeout: 30s

$ echo "# Report" > report.md

cat, grep, echo — that's the API.

Mount via NFS or FUSE and agents read/write with standard POSIX tools. No SDK, no driver, no connection string.

Every write is a commit.

Every write is a commit.

Full history, blame, diff, rollback. Branch from any point. Your knowledge base has the same guarantees as your code.

{

"mcpServers": {

"kiwifs": {

"command": "kiwifs",

"args": ["mcp"]

}}

}

Structured AI interface.

Purpose-built tools for Claude, Cursor, and any MCP-compatible agent. Read, write, search, query — all type-safe.

grep → FTS5 → vector.

grep → FTS5 → vector.

Three tiers that scale with you. Start with zero deps, graduate to BM25, add semantic search when you're ready.

conceptsauth.md

Authentication

[[oauth]][[jwt]]

On this page

Overview

OAuth2

JWT flow

Sessions

A UI you'll actually use.

Block editor, wiki links, backlinks, knowledge graph, Cmd+K search, dark mode. Embedded in the binary — no separate deploy.

One server. Every protocol.

REST, NFS, S3, WebDAV, FUSE, and MCP — all flowing through a single storage layer. Every write gets a git commit, a search index update, and an SSE broadcast.

Loading diagram...

REST API

Web frontend, scripts, CI

NFS

Docker & Kubernetes native mount

S3

Backup, data pipelines

WebDAV

Windows mapped drives

FUSE

Mount remote as local folder

MCP

Claude, Cursor, any AI agent

How it compares

Your agents need persistent, searchable, structured markdown. What do you use?

	KiwiFS	Git repo	Postgres	Obsidian	Chroma
What it is	Markdown filesystem	DIY files + git	Relational database	Local markdown editor	Vector store (OSS)
Agents can write	MCP, REST, cat	echo + manual commit	SQL inserts	Local files only	Upsert embeddings
MCP support	Native (29 tools)	No	No	No	No
Human-readable UI	Built-in web UI	Raw files / GitHub	Needs custom UI	Desktop app	No
Search	FTS (BM25) + vector	grep	SQL LIKE / pg_trgm	Plugins	Vector similarity
Structured queries	DQL over frontmatter	No	SQL (full power)	Plugin (Dataview)	Metadata filters
Versioning	Auto git commits	Manual	WAL / migrations	No	No
Data format	Markdown (portable)	Markdown (portable)	Tables (locked in)	Markdown (portable)	Embeddings (opaque)
Best for	Agent workspace + team wiki	Simple scripts	App backends	Personal notes	OSS semantic search

Ready to give your agents a memory?

Start free on cloud, or self-host the same binary. Open source, BSL-1.1 licensed.

Start Free on Cloud Star on GitHub