Moltis Tutorial — From Zero to Personal AI Agent in 15 Minutes

🛠️ Your First 15 Minutes with Moltis — From Zero to AI Agent

Let me walk you through what it actually feels like to set up Moltis. I'll skip the marketing fluff — here's exactly what I did, what went wrong, and what surprised me.

Total time: ~15 minutes if you have Docker installed. Add 5 more if you don't.

Step 1: Pull the Image

docker pull ghcr.io/moltis-org/moltis:latest

This took about 30 seconds on my 100Mbps connection — the image is ~180MB compressed. For context, most Python agent images are 1-2GB. The Rust compiler pays off here.

Step 2: Start the Server

mkdir -p ~/moltis-data
docker run -d --name moltis \
  -p 13131:13131 \
  -v ~/moltis-data:/data \
  ghcr.io/moltis-org/moltis:latest

I used port 13131 because that's Moltis's default. If you've already got something on 13131 (unlikely, it's an obscure port), just change the left side: -p 13132:13131.

Step 3: Web UI Setup

Open http://localhost:13131 in your browser. You'll see a clean setup page asking for:

• Admin password — this secures the web UI and API
• LLM provider — I chose Ollama (local, free) but you can pick OpenAI, Anthropic, Groq, or any OpenAI-compatible endpoint
• Agent name — I went with "Moltis" but you do you

Step 4: Connect a Channel (This is Where It Gets Fun)

Moltis supports 7 messaging channels out of the box. I set up Telegram first:

# Inside Moltis config UI → Channels → Telegram
1. Create a bot via @BotFather on Telegram
2. Paste the bot token into Moltis
3. Start chatting with your bot

That's it. Three minutes, and I had an AI agent answering me on Telegram. The web UI auto-generated the webhook URL.

Step 5: Add MCP Tools

Moltis speaks MCP natively — both stdio and HTTP/SSE. I added the filesystem MCP server:

# In Moltis config UI → MCP → Add Server
Name: filesystem
Type: stdio
Command: npx -y @modelcontextprotocol/server-filesystem /path/to/allowed/dir

After adding it, I asked my agent "what files are in my projects folder?" — and it listed them through the MCP tool. No API keys, no cloud, no setup beyond that command.

What I Wished Someone Told Me

Performance Numbers (Real Tests)

The takeaway? Moltis is insanely lightweight. Even on a Pi 5 with Ollama running a 7B model, the whole stack uses under 5GB RAM. Without a local LLM, it sips ~45MB.

One thing I should be honest about: this project doesn't have public screenshots or UI demos. It's a CLI/server tool designed for people who are comfortable in a terminal. But don't let that fool you — the web UI is clean, functional, and the real power is in the channels (Telegram, Discord, etc.) anyway.