Comanda

Comanda is an open source command line tool for defining AI workflows as declarative YAML pipelines. Instead of writing glue code around model APIs, developers describe the inputs, models, actions and outputs of each step in a plain text file that can be versioned in Git and run anywhere a terminal runs, from a laptop to CI. It is aimed at developers and operations teams who want repeatable, infrastructure as code style orchestration of LLM work. Comanda runs Claude Code, OpenAI Codex, Gemini CLI and Kimi Code as agents, including full agentic loops with tool access, and can run several in parallel before a synthesis step combines their results. It can generate workflows from natural language, improve them from feedback, and compose them from reusable skills. Codebase indexing covers up to 10,000 source files by default to surface architecture, conventions and design reasoning. A live mode shows a scrollable activity history. Running it as an MCP server turns every workflow in a repository into a callable tool and every skill into a prompt for Claude Code, Codex, Kimi Code and Cursor. Comanda is free to install through Homebrew, Go or prebuilt binaries from GitHub Releases, and no pricing is published. Model usage is billed by the respective providers. A visual workflow builder for macOS, Comanda Canvas, is offered as a download but currently requires a password obtained from the team. The project dates from late 2024, when most agent tooling was SDK first and command line tools were largely limited to single prompts.