Comanda - Declarative AI Pipelines

Comanda defines LLM workflows in YAML and runs them from the command line. Workflows can chain multiple AI models, run steps in parallel, and pipe data through processing stages.

Installation

# macOS
brew install kris-hansen/comanda/comanda

# Or via Go
go install github.com/kris-hansen/comanda@latest

Then configure API keys:

comanda configure

Commands

Generate a Workflow

Create a workflow YAML from natural language:

comanda generate <output.yaml> "<prompt>"

# Examples
comanda generate summarize.yaml "Create a workflow that summarizes text input"
comanda generate review.yaml "Analyze code for bugs, then suggest fixes" -m claude-sonnet-4-20250514

Visualize a Workflow

Display ASCII chart of workflow structure:

comanda chart <workflow.yaml>
comanda chart workflow.yaml --verbose

Shows step relationships, models used, input/output chains, and validity.

Process/Execute a Workflow

Run a workflow file:

comanda process <workflow.yaml>

# With input
cat file.txt | comanda process analyze.yaml
echo "Design a REST API" | comanda process multi-agent.yaml

# Multiple workflows
comanda process step1.yaml step2.yaml step3.yaml

View/Edit Workflows

Workflow files are YAML. Read them directly to understand or modify:

cat workflow.yaml

Workflow YAML Format

Basic Step

step_name:
  input: STDIN | NA | filename | $VARIABLE
  model: gpt-4o | claude-sonnet-4-20250514 | gemini-pro | ollama/llama2 | claude-code | gemini-cli
  action: "Instruction for the model"
  output: STDOUT | filename | $VARIABLE

Parallel Execution

parallel-process:
  analysis-one:
    input: STDIN
    model: claude-sonnet-4-20250514
    action: "Analyze for security issues"
    output: $SECURITY

  analysis-two:
    input: STDIN
    model: gpt-4o
    action: "Analyze for performance"
    output: $PERF

Chained Steps

extract:
  input: document.pdf
  model: gpt-4o
  action: "Extract key points"
  output: $POINTS

summarize:
  input: $POINTS
  model: claude-sonnet-4-20250514
  action: "Create executive summary"
  output: STDOUT

Generate + Process (Meta-workflows)

create_workflow:
  input: NA
  generate:
    model: gpt-4o
    action: "Create a workflow that analyzes sentiment"
    output: generated.yaml

run_it:
  input: NA
  process:
    workflow_file: generated.yaml

Available Models

Run comanda configure to set up API keys. Common models:

| Provider | Models | |----------|--------| | OpenAI | gpt-4o, gpt-4o-mini, o1, o1-mini | | Anthropic | claude-sonnet-4-20250514, claude-opus-4-20250514 | | Google | gemini-pro, gemini-flash | | Ollama | ollama/llama2, ollama/mistral, etc. | | Agentic | claude-code, gemini-cli, openai-codex |

Examples Location

See ~/clawd/comanda/examples/ for workflow samples:

• agentic-loop/ - Autonomous agent patterns
• claude-code/ - Claude Code integration
• gemini-cli/ - Gemini CLI workflows
• document-processing/ - PDF, text extraction
• database-connections/ - DB query workflows

Troubleshooting

• "model not configured": Run comanda configure to add API keys
• Workflow validation errors: Use comanda chart workflow.yaml to visualize and check validity
• Debug mode: Add --debug flag for verbose logging