kiwifs/kiwifs

KiwiFS – a markdown‑first knowledge store for AI agents and teams

What it is – A single‑binary Go server that treats a directory of Markdown files as a live, searchable, version‑controlled knowledge base. Files are the source of truth; the system adds Git‑style audit trails, full‑text + vector search, and a web UI, while exposing the data through many protocols (REST, MCP, NFS, S3, WebDAV, FUSE) so both humans and autonomous agents can read and write.

Why it matters for AI – Modern LLM‑powered agents need a place to persist context that is both machine‑readable (embeddings, structured front‑matter) and human‑readable (Markdown). KiwiFS lets agents:

• write Markdown via simple cat‑style commands, HTTP PUT, or the MCP protocol,
• retrieve relevant pages with BM25 or semantic vector similarity,
• query structured metadata with a SQL‑like DQL language,
• see every change as a Git commit (blame, diff, point‑in‑time restore),
• receive live updates through Server‑Sent Events.

Key features (as listed in the README)

Feature	What you get
62 MCP tools	Native commands for Claude, Cursor and any MCP client.
Embedded Web UI	Block editor, wiki‑link backlinks, knowledge‑graph view, dark mode – all compiled into the binary via go:embed.
Full‑text + vector search	BM25 via SQLite FTS5; plug‑in vector back‑ends (OpenAI, Ollama, ONNX, Cohere, Qdrant, pgvector, Pinecone).
Git versioning	Every write becomes an atomic Git commit; you can git blame, diff, or restore any revision.
DQL (Data Query Language)	SQL‑style queries over Markdown front‑matter (TABLE, WHERE, GROUP BY, etc.).
Six access protocols	REST, MCP, NFS, S3, WebDAV, FUSE – all routed through a single storage layer.
19 importers	Pull data from Postgres, MySQL, MongoDB, Notion, CSV, Obsidian, …
Content health analytics	Detect stale pages, broken links, orphaned notes, contradictions; trust‑ranked search.
Multi‑space	Host multiple isolated workspaces, each with its own Git repo and index.
Webhooks & schema validation	Trigger Slack/CI on writes, enforce JSON‑Schema on incoming Markdown.
Embeddable Go library	Use pkg/kiwi inside your own Go applications.

Typical users

• AI‑agent developers who need a writable knowledge base that agents can query and update program‑matically.
• Teams replacing Confluence/Notion with a self‑hosted, Markdown‑centric wiki that still offers a modern UI.
• Compliance‑heavy organisations that require immutable audit trails (Git commits) for every change.
• DevOps / SRE teams that want runbooks that agents can automatically enrich after incidents.

How to get started (from the README)

# Install (brew, curl script, or go install)
brew install kiwifs/tap/kiwifs
# or
curl -fsSL https://raw.githubusercontent.com/kiwifs/kiwifs/main/install.sh | sh
# or
go install github.com/kiwifs/kiwifs@latest

# Initialise a workspace
kiwifs init --template knowledge --root ./knowledge

# Run the server (REST API on :3333, UI at http://localhost:3333)
kiwifs serve --root ./knowledge

Docker alternative:

docker run -p 3333:3333 -v ./knowledge:/data ameliaanhlam/kiwifs

Agents can write via a simple HTTP request:

curl -X PUT 'localhost:3333/api/kiwi/file?path=pages/auth.md' \
  -H "X-Actor: my-agent" \
  -d "# Authentication\n\nOAuth2 + JWT..."

Architecture snapshot (text diagram from the README)

• Single Go binary embeds the Web UI (shadcn/ui, CodeMirror, Sigma.js).
• Access layer exposes REST, MCP, NFS, S3, WebDAV, FUSE.
• Core handles storage, Git versioning, SQLite FTS5, optional vector store, file‑watcher, SSE events, schema linting.
• Filesystem back‑ends include local disk, NFS, EFS, JuiceFS, FUSE‑S3.

Extensibility

• Plug‑in vector providers (OpenAI, Ollama, ONNX, etc.) via [search.vector] in config.toml.
• Add custom importers or export pipelines (JSONL/CSV with embeddings).
• Use the Go library directly in other programs.

License – Business Source License 1.1 (free for self‑hosted use; commercial SaaS offering requires a separate license). After four years each release re‑licenses to Apache 2.0.

Where to learn more – Docs site (https://docs.kiwifs.com), API spec, examples, FAQ, roadmap, and architecture docs are all linked from the README.

---

All information above is taken directly from the repository’s README; no additional features have been inferred.