Tabularium is an AI-oriented markdown document store with full-text search, a real directory tree, and several ways to work with the same data: web UI, CLI, REST, JSON-RPC, MCP, or the Rust library.
What It Does
- Stores documents in SQLite with Tantivy-backed search.
- Lets you browse, edit, append, search, and export documents as a tree.
- Serves the same data through an embedded web UI, HTTP APIs, MCP, and
tb. - Supports meeting and chat-style document workflows for humans and machine spirits alike.
Workspace
tabularium: core library crate.tabularium-server: HTTP server with embedded web UI, REST, JSON-RPC, and optional MCP.tabularium-cli:tbcommand-line client with one-shot commands, interactive shell, and chat flows.ui: Vite/React frontend embedded into the server build.
Linux installation
|
Configuration file: /etc/tabularium/config.toml
Quick Start
With the example config, this starts:
- web UI, REST, and JSON-RPC at
http://127.0.0.1:3050 - MCP at
http://127.0.0.1:3031/mcp
In another terminal:
just run uses config.toml when present and falls back to config.toml.example for a fresh clone.
Install (Homebrew)
# add tap
# install
Apple Silicon only (arm64 / aarch64-apple-darwin). Intel macOS (x86_64-apple-darwin) artifacts are not published yet.
This installs:
tabularium-serveras a Homebrew service targettbas a regular CLI binary inbin(not a service)
Start and manage the server with:
Docs
AI Agents
When an AI agent uses Tabularium as shared project memory, point it at the MCP endpoint and keep it on the MCP tool surface. The practical setup and meeting workflow live in docs/AI-Agents.md.
Human chat and meeting usage across the Web UI and tb chat is described in docs/chat-mode.md.
Development
Safety notes
Tabularium is an experimental AI-blackbox project, mean no human code review
is performed. Additional safety measures may be required, such as sandboxing,
regular data backups, and monitoring for unusual activity.
License
Apache-2.0. See LICENSE.