CLI Reference

The chia console script (chia/cli/main.py) manages clusters, visualizes flows, and wraps Ray job management.

chia <command> [options]

Cluster management

chia up <config.yaml>

Bring up a Chia/Ray cluster from a YAML config.

Flag

Meaning

-y, --yes

Skip the confirmation prompt.

-v, --verbose

DEBUG logging.

--dry-run

Print the plan and generated scripts without executing.

--add

Only add new nodes to an existing cluster (skip ones already up).

chia down <config.yaml>

Tear down the cluster. Flags: -y/--yes, -v/--verbose.

Visualization

chia viz <source.py>

Render the @ChiaFunction call graph of a flow statically from source.

Flag

Meaning

--func NAME

Orchestrator function (auto-detected if omitted).

--format {svg,png,pdf}

Output format (default svg).

--output-dir DIR

Output directory (default: alongside the source).

chia viz-profile <log...>

Visualize a ChiaProfileCollector JSONL log. See the Profiling guide for how to record one.

Flag

Meaning

--format {svg,png,pdf,html,table}

Output format (default svg); table aggregates exec time to CSV.

--output-dir DIR

Output dir (ignored for table).

--output PATH

(table) CSV path; default stdout.

--funcs ...

(table) function names, or a file of names.

--run N

Run index to visualize (default: last).

--gap-threshold SEC

Gap that segments runs (default 600).

--x-scale F

Horizontal inches per second of wall-clock (default 0.5).

Jobs

chia job <subcommand> — submit, status, logs, list, and delete are pass-throughs to ray job (so --help shows Ray’s own help). stop is custom:

chia job stop <job_id>

By default a thin wrapper over ray job stop. With --kill-tracked-pids, it first asks the PID registry actor to kill every tracked subprocess (SIGTERM, then SIGKILL for any that linger past the grace period), then stops the job. The kill returns as soon as the processes have actually exited, so it normally completes in well under the grace period.

Flag

Meaning

--kill-tracked-pids

Kill tracked subprocesses (via the PID registry) before stopping the job. Off by default.

--grace-period SEC

Seconds to wait for each tracked subprocess to exit after SIGTERM before escalating to SIGKILL (only used with --kill-tracked-pids; default 25).

FireSim

FPGA build/run orchestration:

  • chia firesim-build <config.yaml> --recipe NAME [--instance-type ...]

  • chia firesim-run <config.yaml> [--hw-config ... --workload ... | --suite ...] [...]

  • chia firesim-upload-workload [config.yaml] --marshal-config ... --images-dir ... --suite-name ...

  • chia firesim-cleanup <config.yaml> — terminate orphaned chia EC2 instances.

Run chia <command> --help for the full, current option list.