Declarative AI workflows for the command line.

Comanda turns repeatable AI work into YAML pipelines you can run, review, and version control. Use it to generate workflows from natural language, run multi-model pipelines, orchestrate agentic loops, process files, call tools, and wire Claude Code, Gemini CLI, OpenAI Codex, Kimi Code, and API models together.

For the full guide, feature tour, and copy-ready workflow templates, start at comanda.sh:

• Features
• Templates
• GitHub examples

brew install kris-hansen/comanda/comanda

Or install with Go:

go install github.com/kris-hansen/comanda@latest

See GitHub Releases for prebuilt binaries.

comanda configure
comanda generate workflow.yaml "review this code for bugs"
comanda process workflow.yaml

Pipe input through a workflow:

cat main.go | comanda process workflow.yaml

Inspect or iterate on a workflow:

comanda chart workflow.yaml
comanda improve workflow.yaml "Add security findings and suggested fixes"

summarize:
  input: STDIN
  model: gpt-4o
  action: "Summarize the input in three bullets."
  output: STDOUT

Run it:

cat notes.md | comanda process summarize.yaml

• Multi-agent reviews with Claude Code, Gemini CLI, OpenAI Codex, Kimi Code, and API models
• Agentic loops that iterate until work is complete
• File, URL, image, PDF, database, and batch-processing workflows
• Tool-enabled workflows with explicit command allowlists
• Codebase indexes for persistent project context
• Git worktree workflows for parallel isolated implementation
• Server-backed workflows callable over HTTP
• MCP server mode exposing workflows as tools and skills as prompts

Most examples and walkthroughs live on comanda.sh:

• Browse workflow templates
• Explore all features
• Local examples directory
• Multi-agent patterns
• Agentic loops
• Tool use
• Server API
• MCP server

Run comanda as an MCP server so agent clients (Claude Code, Kimi Code, Codex) can call workflows as tools and use skills as prompts:

comanda mcp

Each discovered workflow file (from ~/.comanda/workflows/, .comanda/workflows/, --dir, or --workflow) becomes one MCP tool; each skill becomes an MCP prompt. See docs/mcp-server.md and examples/mcp/ for client setup and a walkthrough.

make deps
make build
make test

MIT