dag run

Execute an AI pipeline. Pipelines can be defined inline with --pipeline or loaded from a .dag.json workflow file.

Synopsis

bash

dag run --pipeline "<node> | <node> | ..." [--input key=value] [--watch] [--stdin]
dag run <workflow.dag.json> [--input key=value] [--watch] [--stdin]

Inline Pipeline

Use the --pipeline flag to define a pipeline as a pipe-separated string. Each segment is a node type. The special input node injects your --input values and text-output prints the result to stdout.

bash

dag run \
  --pipeline "input | llm-text-anthropic | text-output" \
  --input text="Hello, what is 2 + 2?"

Example output:

text

2 + 2 equals 4.

File Mode

Pass a .dag.json file as the first positional argument to run a saved workflow. The file must follow the DAG workflow file format.

bash

dag run workflow.dag.json \
  --input text="Summarise the following article: ..."

Workflow files can be created with dag build, dag describe, or dag from-mermaid.

Watch Mode

Add --watch to keep the pipeline alive. Each line you type is treated as a new text input and the pipeline runs again.

bash

dag run \
  --pipeline "input | llm-text-anthropic | text-output" \
  --watch

text

> Translate "hello" to Spanish.
hola

> Translate "goodbye" to Spanish.
adiós

> ^C

Reading from stdin

Use --stdin to pipe text directly into the pipeline. This is equivalent to --input text=<stdin>.

bash

echo "What is the capital of France?" | dag run \
  --pipeline "input | llm-text-anthropic | text-output" \
  --stdin

Flags

Flag	Description
--pipeline <string>	Inline pipeline definition. Nodes separated by \|, data flows left to right.
--input <key=value>	Pass input values to the pipeline. Repeat the flag for multiple keys: --input text="..." --input lang="en".
--watch	Re-run the pipeline on each line read from stdin. Useful for interactive sessions.
--stdin	Read input text from stdin instead of --input text=. Equivalent to piping text into the pipeline.