awpak-ai-cmd-client

awpak-ai-cmd-client is a command-line client for executing AI workflows defined as graphs using the awpak-ai library.
This tool allows anyone, even without programming experience, to create and run AI-driven workflows.

Features

• Execute workflows (graphs) defined in JSON.
• Interactive chat mode with LLMs.
• Detailed trace options to control execution output.
• Support for streaming agent responses.

Installation

You can install it from source using Cargo:

cargo install awpak-ai-cmd-client

Usage

awpak-ai-cmd-client --path="/path_to_graph.json" [OPTIONS]

Required Arguments

• --path <FILE> Path to the JSON file containing the graph definition.

Optional Arguments

• --input <INPUT> Graph input value. Ignored when using --chat.

• --trace <TRACE_OPTIONS> Comma-separated trace options to control output visibility. Default: graph_output_ok,graph_output_err.

  Available trace options:

  • graph_input → Prints the graph input
  • graph_output_ok → Prints the graph output if execution succeeds
  • graph_output_err → Prints the graph output if execution fails
  • agent_prompt → Prints the prompt for Agent nodes
  • agent_stream → Shows streaming output from Agent nodes (if enabled)
  • agent_sync → Shows synchronous output from Agent nodes
  • agent_tool_call → Shows MCP tool calls made by Agent nodes
  • agent_tool_result → Shows the result of MCP tool calls
  • command_and_args → Shows the command and arguments for Command nodes
  • command_result → Shows the result of Command nodes
  • web_client_request → Shows the URL and method for WebClient nodes
  • web_client_request_body → Shows the body of WebClient requests
  • web_client_request_headers → Shows headers of WebClient requests
  • web_client_request_query_params → Shows query parameters of WebClient requests
  • web_client_response → Shows version and status code of WebClient responses
  • web_client_response_headers → Shows headers of WebClient responses
  • web_client_response_body → Shows body of WebClient responses
  • node_destination → Shows each node executed and the chosen destination
  • node_execution → Shows node IDs before execution
  • node_output → Shows output of each node

• --chat Start interactive chat mode. Input is read from a prompt instead of --input.

Example

Running a Graph with Static Input

awpak-ai-cmd-client --path="./graph.json" --input="Hello world" --trace="agent_stream"

Running in Chat Mode

awpak-ai-cmd-client --path="./graph.json" --trace="agent_stream" --chat

This starts an interactive shell where each message you type is sent as graph input.

Related Projects

• awpak-ai: Core Rust library for building and executing AI workflows.
• awpak-ai-graph-builder: Web-based tool to easily create workflow graphs.

License

This project is licensed under the MIT license.