OpenFang Tutorial — 5 Minutes to Your First Autonomous Agent

Getting Started with OpenFang — 5 Minutes to Your First Autonomous Agent

I've been through enough "quick start" guides that turned into hour-long debugging sessions. So let me walk you through setting up OpenFang the way I wish someone had shown me — including the mistakes I made so you don't have to repeat them.

Installation — One Command, No Surprises

OpenFang runs on macOS, Linux, and Windows. The binary is ~32 MB. Here's the install command:

curl -fsSL https://openfang.sh/install | sh

That's it. No Docker needed, no pip, no npm, no Python version drama. Time elapsed: ~10 seconds on a decent connection.

Initialize — This Is Where You Set Up LLM Access

openfang init

The init command walks you through selecting an LLM provider. OpenFang supports 27 providers and 123+ models, including OpenAI, Anthropic, Groq, Together AI, Ollama (for local models), and many more.

Don't overthink this — start with whatever API key you already have. I used Groq because it's free and fast. You can switch providers later from the dashboard.

Start the Daemon — Your Agents Go Live

openfang start

This starts the OpenFang daemon in the background. You'll see log output showing the daemon PID and the dashboard URL.

Open your browser and go to http://localhost:4200. If you see the OpenFang dashboard, you're in business.

Activate Your First Hand — Get Real Work Done

Here's the fun part. Activate the Researcher Hand to start getting autonomous research reports:

openfang hand activate researcher

Now chat with it:

openfang chat researcher
> "What are the emerging trends in AI agent frameworks this quarter?"

Use Docker Instead? Sure, Here's How

If you prefer Docker, OpenFang runs in containers too. The docker-compose setup from Run This Ai exposes it on port 8080:

docker pull fliva/openfang:latest
docker run -d --name openfang -p 8080:8080 -v ./data/openfang:/data fliva/openfang:latest

What I Learned After a Week

A few things that surprised me:

1. The Researcher Hand is insanely good. I asked it to analyze my competitors' product launches over the past 3 months. It cross-referenced 12 sources, evaluated credibility using CRAAP criteria, and produced a 4-page report with APA citations. Took about 4 minutes. That would have been a full afternoon of manual work.

2. Memory is persistent. OpenFang uses SQLite with vector embeddings for canonical sessions. If you ask the same agent a follow-up question 3 days later, it remembers context. This is surprisingly rare in agent frameworks.

3. MCP is built-in. OpenFang ships with 25 MCP templates and can act as both an MCP client and server. If you're running other MCP-compatible tools, they can talk to each other out of the box.

4. Cold start is ~180ms. Coming from LangGraph where I'd wait 2.5 seconds for every agent spawn, this felt like magic.

Final Tips From My Experience

• Start with the Researcher Hand — it's the most immediately useful and requires no configuration beyond an LLM provider.
• Set up a schedule — openfang hand schedule researcher --cron "0 8 * * 1-5" makes it run every weekday at 8 AM automatically.
• Connect a channel — openfang channel add telegram --token YOUR_BOT_TOKEN lets agents deliver reports to your chat.
• Pin your version — OpenFang is pre-1.0. Use openfang version pin 0.6.9 to avoid surprise breaking changes.