CLI reference¶

The agentic command is installed by the agentic-workflows-v2 package:

pip install -e "./agentic-workflows-v2[server,langchain]"
agentic --help

Run commands from the repository root or pass explicit paths. Relative workflow paths, input files, output files, and checkpoint directories are resolved from the current directory.

Command map¶

Command	Purpose
`agentic run`	Run a named or path-based YAML workflow
`agentic compare`	Run one workflow through multiple execution adapters
`agentic orchestrate`	Placeholder: prints a notice and exits 1 (dynamic orchestration is not exposed through the CLI yet)
`agentic resume`	Read a saved execution checkpoint and report changed files
`agentic list`	List workflows, agents, tools, or execution adapters
`agentic validate`	Validate a workflow without running it
`agentic serve`	Start the FastAPI server and packaged UI
`agentic version`	Print the installed `agentic-workflows-v2` version
`agentic devex`	Run development checks

`agentic run`¶

agentic run [OPTIONS] WORKFLOW

WORKFLOW is either a built-in workflow name such as code_review or a path to a YAML file.

Option	Type	Default	Meaning
`--input`, `-i`	path	none	JSON file containing workflow inputs
`--output`, `-o`	path	none	Write the run result as JSON
`--dry-run`	flag	off	Validate and print the execution plan without running steps
`--verbose`, `-v`	flag	off	Print step-level details
`--adapter`, `-a`	text	`langchain`	Use `langchain` or `native`

The input option accepts a file path:

agentic run code_review --input review-input.json --adapter native

--output writes a JSON summary with the same shape for both adapters (it is not the full WorkflowResult model, so there is no final_output or overall_status key):

Key	Contents
`workflow_name`	Workflow that ran
`status`	Lowercase overall status, for example `success` or `failed`
`outputs`	The workflow's declared outputs
`steps`	Map of step name to `status`, `outputs`, and `error`
`errors`	List of error messages
`elapsed_seconds`	Wall-clock duration

The command exits non-zero when validation fails or the workflow cannot be run. A workflow result whose overall status is failed is also reported as a failed command.

`agentic compare`¶

agentic compare [OPTIONS] WORKFLOW

Option	Type	Default	Meaning
`--input`, `-i`	path	required	JSON input file used for every adapter
`--adapters`	comma-separated text	`native,langchain`	Adapter names to run

agentic compare code_review \
  --input review-input.json \
  --adapters native,langchain

The command prints status, step count, and elapsed time for each adapter. It exits non-zero if any adapter fails, because a one-sided result is not a valid comparison.

`agentic orchestrate`¶

agentic orchestrate [OPTIONS] TASK

TASK is a text description. The command is currently a stub: it prints that dynamic orchestration is not exposed through the default LangGraph workflow path yet, points to agentic run <workflow>, and exits with status 1 without planning or running anything, whatever the options or AGENTIC_NO_LLM setting. The options below are accepted but unused. The orchestrator itself is available from Python (see the Pattern catalog).

Option	Type	Default	Meaning
`--max-parallel`, `--max-steps`	integer	`3`	Maximum number of steps dispatched at the same time
`--verbose`, `-v`	flag	off	Print planning and execution details

Use agentic run <workflow> for YAML-defined workflows.

`agentic resume`¶

agentic resume [OPTIONS] NAME

NAME is the checkpoint file stem.

Option	Type	Default	Meaning
`--checkpoint-dir`	path	`.agentic_checkpoints`	Directory containing checkpoint JSON files
`--fork`	text	none	After loading, branch an in-memory fork with this name and print its run id

agentic resume my_run --checkpoint-dir ./checkpoints
agentic resume my_run --checkpoint-dir ./checkpoints --fork experiment_a

The command restores an ExecutionContext, compares tracked files with their saved state, and reports which files changed. It does not restore a model's hidden conversation state. --fork only calls ExecutionContext.fork_session and prints the new run id; the CLI does not write a checkpoint for the fork or continue the run. Exit status is 1 when the checkpoint file is missing or cannot be restored.

`agentic list`¶

agentic list [COMPONENT_TYPE]

COMPONENT_TYPE is one of:

workflows (default)
agents
tools
adapters

agentic list
agentic list tools

`agentic validate`¶

agentic validate [OPTIONS] WORKFLOW

Option	Type	Default	Meaning
`--verbose`, `-v`	flag	off	Print workflow counts and the execution plan

Validation checks YAML syntax, the workflow schema, missing dependencies, cycles, and LangGraph compilation. The compilation check requires the langchain extra even when you intend to execute the workflow with the native adapter.

agentic validate code_review --verbose
agentic validate ./workflows/custom.yaml

`agentic serve`¶

agentic serve [OPTIONS]

Option	Type	Default	Meaning
`--port`, `-p`	integer	`8000`	HTTP port
`--dev`	flag	off	Enable server auto-reload
`--no-open`	flag	off	Do not open a browser

agentic serve --port 8010 --dev --no-open

The repository's just dev launcher uses backend port 8010 and Vite port 5173. The standalone CLI uses port 8000 unless you override it.

`agentic version`¶

agentic version

Prints one line, agentic-workflows-v2 version <version>; it reports no other environment information.

`agentic devex`¶

Port checks¶

agentic devex port-guard --backend-port 8010 --frontend-port 5173

The command's built-in defaults are 8012 and 5174, which differ from just dev. Pass the ports explicitly when checking the standard development launcher.

Workspace tests¶

agentic devex workspace-test-runner
agentic devex workspace-test-runner --coverage
agentic devex workspace-test-runner --package agentic-workflows-v2

Option	Default	Meaning
`--skip-integration` / `--no-skip-integration`	skip	Exclude or include tests marked `integration`
`--coverage` / `--no-coverage`	no coverage	Add coverage reporting
`--package`	all	Limit the run to `agentic-workflows-v2`, `agentic-v2-eval`, or `tools`

Workflow linter¶

agentic devex workflow-linter code_review
agentic devex workflow-linter ./workflow.yaml --strict

--strict treats warnings as violations.

Environment behavior¶

Settings are read from process environment variables and the repository .env file. See Configuration.
AGENTIC_NO_LLM=1 installs fixed-response model substitutes for both execution engines. It does not install missing optional packages.
Named YAML runs default to langchain; runtime-generated DAGs use the native engine.
Paths are relative to the shell's current directory.

Getting help¶

Every group and command supports --help:

agentic --help
agentic run --help
agentic devex workspace-test-runner --help

Typer's generated help is the source of truth for installed command options. This page records the command behavior in the current repository.

CLI reference¶

Command map¶

agentic run¶

agentic compare¶

agentic orchestrate¶

agentic resume¶

agentic list¶

agentic validate¶

agentic serve¶

agentic version¶

agentic devex¶