mdd

MDD's terminal coding assistant. Chat with an AI agent that reads and edits files, runs shell/git, and streams its work in a polished terminal UI — multi-provider (Anthropic + OpenAI-compatible, including 9router), with confirmation before any change.

Features

• 🤖 Agentic loop — reads, searches, edits files and runs shell & git to complete tasks
• 🧩 Skills — reusable SKILL.md instruction bundles the agent loads on demand
• 🔌 Multi-provider — Anthropic, OpenAI, or any OpenAI-compatible endpoint (9router)
• 📊 Live token + estimated-cost meter in the status bar
• 🖥️ Fullscreen TUI — gradient header, live tool status & timing, markdown replies, themes
• 🔒 Safe by default — every file/shell/git change asks for confirmation first
• ⚡ Switch on the fly — /model, /provider, /theme without restarting
• 📜 History persists — your conversation prints back to the terminal on exit

Quick start (for MDD teammates)

1. Install (needs Node.js 20+):

npm install -g mdd-cli

2. Run it — the first time, it walks you through setup automatically:

mdd

A short wizard walks you through it. For the company 9router, just:

1. Choose 1) 9router
2. Paste your API key (ask Daffa for it)

That's it — the endpoint and a good default model are set for you. Pick a different model any time with /models, then start chatting. 🎉

Update later:

npm install -g mdd-cli@latest

Everyday use

mdd                                     # interactive chat (fullscreen TUI)
mdd "add a health check and run tests"  # one-shot: answer and exit
mdd --yes "reformat src/**/*.ts"        # auto-approve file/shell changes (careful!)

Mutating tools (write/edit files, shell, git) ask for confirmation. Read-only tools run silently. Your conversation prints to the terminal when you exit, so it stays in your scrollback.

In-chat commands

Type these inside the chat (a leading / means it's a command, not a prompt):

/models            pick a model with ↑/↓ and Enter
/model [id]        show or switch the model directly
/provider <name>   switch provider: anthropic | openai
/theme [name]      switch theme: neon | ocean | mono
/help              list commands
/exit              quit (or press Ctrl-C)

PageUp / PageDown scroll earlier messages. The status bar shows the active provider · model, working directory, and git branch.

9router (and any OpenAI-compatible endpoint)

mdd reaches 9router through the openai provider — you just point it at the base URL and use a 9router model id. The setup wizard stores this for you, or per-invocation:

mdd --provider openai \
    --base-url http://192.168.7.8:20128/v1 \
    --model cc/claude-sonnet-5 \
    "explain this repo"

mdd models lists the known 9router (cc/*) ids. Base-URL precedence: --base-url flag → OPENAI_BASE_URL env → stored config → OpenAI's default.

The routed model must support streamed OpenAI function calling for tools to fire. If a model replies with text but never runs a tool, that backend doesn't support it.

Skills

Skills are small, reusable bundles of instructions the agent can load on demand. Each skill is a folder with a SKILL.md file; the agent sees a short list of available skills in its system prompt and pulls a skill's full instructions into context (via the use_skill tool) only when a task matches.

Skills are discovered from two places:

• Project skills: <repo>/.mdd/skills/<name>/SKILL.md — checked into the repo, shared with your team.
• Personal skills: ~/.config/mdd/skills/<name>/SKILL.md — just for you.

A project skill overrides a personal skill with the same name.

A minimal SKILL.md:

---
name: deploy
description: How to release the app to staging
---

1. Run `npm run build`.
2. Push the image with `./scripts/deploy.sh staging`.
3. Verify the health check at https://staging.example.com/healthz.

The frontmatter (name, description) is optional — with no frontmatter, the folder name becomes the skill name and the whole file is the instruction body. Skills are read-only and load without a confirmation prompt.

Config

Stored at ~/.config/mdd/config.json (mode 0600). Re-run setup any time with mdd auth login. Environment overrides: ANTHROPIC_API_KEY, OPENAI_API_KEY, OPENAI_BASE_URL.

Roadmap

GitLab and Odoo tools (read-only first), central secrets, richer skills (bundled scripts/resources, a /skills command).