Mosfet
Early-stage · work in progress

Your own AI agent, on your own machine.

Mosfet is an LLM-powered agent you extend with skills, tools, MCP servers and workflows. Drop in more building blocks, and it can do more.

bash <(curl -fsSL https://mosfet.shahtech.info)
InstallView on GitHub

Features

Everything it can do today.

  • Chat where you are

    Talk to the agent from your terminal or other communication channels. Each channel picks up its last conversation right where you left off.

  • Skills and workflows

    The agent can create skills for itself and reuse them whenever they apply. Save multi-step procedures as workflows and run them again whenever you need.

  • Connects to MCP servers

    Connect the agent to MCP servers so it can use their tools.

  • Works through it, visibly

    It works through multiple steps on its own before giving you a final answer, and shows what it's doing as it goes, like loading a skill or calling a tool.

  • Safe by default

    Safe, read-only actions run automatically, and it asks your permission before doing anything else. Messages are checked for sensitive personal information on the way to the LLM and on the way back.

  • Your model, your machine

    Point it at whichever LLM provider and model you want, hosted or local, and change it anytime. Install once and run it from anywhere on your machine.

See it in action

Mosfet chatting in a terminal: the user's question, the agent's tool activity, and its answer.
CLI
Mosfet replying to a direct message in Discord.
Discord

Install

Pick the way that suits you.

  1. bash <(curl -fsSL https://mosfet.shahtech.info)

  2. This clones the latest Mosfet release, sets up an isolated Python environment, and installs a mosfet command on your PATH.

  3. Then set it up and start chatting:

    mosfet config
mosfet

  1. Create a docker-compose.yml in the project directory:

    services:
  mosfet-app:
    image: busycaesar/mosfet:latest
    volumes:
      - ./mosfet.config.json:/app/mosfet.config.json
      - ./.env:/app/.env

  2. Create mosfet.config.json and .env in the same directory:

    echo '{}' > mosfet.config.json
touch .env

  3. Start the container:

    docker compose up -d

  4. Set it up and start chatting:

    docker compose exec -it mosfet-app mosfet config
docker compose exec -it mosfet-app mosfet

  1. Clone the repo:

    git clone git@github.com:busycaesar/Mosfet.git
cd Mosfet

  2. Create a virtual environment and install dependencies:

    python3 -m venv .venv
source .venv/bin/activate
pip install -r req.txt

  3. Set it up and start chatting:

    python3 src/main.py config
python3 src/main.py

FAQ

Is it free?

Yes. Mosfet is open source under the MIT License.

Which LLMs can it use?

Whichever LLM provider and model you choose, hosted or local. You can change it anytime by running mosfet config.

Where does it run?

On your own machine. Install it with the install script, run it with Docker, or set it up manually from source.

How do I talk to it?

Through your terminal or other communication channels. Each channel resumes its last conversation automatically.

How do I extend it?

Add skills, tools, MCP servers and workflows. Skills describe a capability or knowledge it can draw on, tools let it take action, MCP servers connect it to external systems, and workflows chain these into a saved, repeatable procedure.

Is it ready to use?

It's early-stage and a work in progress. You can install it and try it today with any of the methods above.