GitHub - hybridgroup/skynet: Skynet is a lightweight command line interface for LLMs to control real robots and drones with MCP using Bash.

Skynet

🔺 What is Skynet

Skynet is a lightweight command line interface for LLMs to control real robots and drones using Bash.

It can run entirely on your own machine, or can be configured to run using remote services.

Skynet provides an MCP host that calls a model server such as Docker Model Runner to allow Large Language Models (LLM), Small Language Models (SLM), and Tiny Language Models (TLM), to control physical devices that provide an MCP server interface. See ROBOTS.md for some known devices with MCP servers.

It supports multiple step interactions when using models such as Qwen 2.5 Instruct. See MODELS.md for some models that are known to support multistep actions.

🔺 How Skynet works

Skynet is written in Bash using Osprey.

Architecture

flowchart TD
subgraph Skynet
    A(["Chat interface"])
    A <---> C["Chat completions"]
    A <---> T["Tool calls"]
    C <--> T
end
subgraph "Docker Model Runner"
  M["Qwen 2.5"]
end
subgraph "MCP Server 1"
  D["Drone"]
end
subgraph "MCP Server 2"
  R["Robot"]
end
C <-- http --> M
T <-- http --> D
T <-- http --> R

User flow

sequenceDiagram
        actor Human
        actor Skynet
        actor Model
        actor MCP Server
        actor Robot
        loop entering commands
            Human->>Skynet: type command
            Skynet->>Model: send message
            Model->>Skynet: send response
            loop while tool calls needed
                Skynet->>MCP Server: call tool
                MCP Server-->>Robot: call device API
                Robot-->>MCP Server: device API result
                MCP Server->>Skynet: send results from tool
                Skynet->>Skynet: add results to message history
                Skynet->>Model: send updated message
                Model->>Skynet: send response
            end
            break no tool calls needed
                Skynet->>Human: display result
            end
        end

🔺 Installing Skynet

See INSTALL.md for information on how to install Skynet.

🔺 Running Skynet

$ ./skynet.sh -h
Usage: ./skynet.sh [OPTIONS]
Options:
 -h, --help           Display this help message
 -v, --version        Display version
 -d, --debug          Enable debug mode
 -m, --model          Model to use
 -p, --pull-model     Pull latest model
 -s, --model-server   Model server to use
 -i, --instructions   System instructions to use
 -t, --temperature    Temperature for model
 -r, --robot-server   Robot servers to use

You must have any robot MCP servers already running in order to run Skynet. See ROBOTS.md for a list of some known physical devices with MCP server interfaces.

Then run the skynet.sh command, including any flags that may be needed.

Additional command options

You can set the following flags or environment variables to change Skynet.

🔺 -d, --debug

Set debug mode. Defaults to false

🔺 -m, --model

Set the model to use. Defaults to ai/qwen2.5:latest

🔺 -p, --pull-model

Should Docker Model Runner pull the latest model? Defaults to true

🔺 -s, --model-server

Model server to use. Defaults to Docker Model Runner http://localhost:12434/engines/llama.cpp/v1

🔺 -i, --instructions or SYSTEM_INSTRUCTION

System instructions to use. Defaults to:

You are a robot.
You have tools that actually call devices in the physical world that you are connected to.
Use your tools to respond to human requests.
Keep your responses short and to the point.

🔺 -t, --temperature

Set the temperature for the model. Defaults to 0.0

🔺 -r, --robot-server

Set the MCP server to use. Allows multiple addresses separated by comma. Note that tool names must be unique per MCP server. Defaults to http://localhost:9090

🔺 OSPREY_INSTALL

Env var to set the directory in which to find Osprey. Defaults to your same directory as Skynet.

🔺 Warning

You know the legends. Usage of this software is at your own risk.