Skip to content

Command-line tools

MAS-Lab ships two CLIs. mas-ctl runs agents and MAS manifests. mas-lab runs experiments, telemetry, and workspace bootstrap.

mas-runtime is a library (no CLI). Use mas-ctl chat / run-mas.

CLI Package Use for
mas-ctl ctl/ Chat, TUI, compile, compose, validate, run-mas
mas-lab lab/ init, benchmarks, pipelines, telemetry, plots
mas-ctl --help / mas-ctl COMMAND --help — Always-current flag list

Complete option tables: mas-ctl.md. Workspace file: config.yaml reference. Machine run logs: observability.md (events.jsonl).


What goes where

Stream Consumer How you get it
stdout Humans Conversation (You: / Agent:)
stderr exchange log Humans --trace or mas_ctl.trace in config.yaml
events.jsonl Machines --events, overlay, or manifest spec.observability
mas-lab telemetry Machines Post-run inspect / export of events.jsonl

Do not score experiments from the exchange log.


mas-ctl (day to day)

mas-ctl chat agent.yaml -q "What is 2+2?"
mas-ctl run-mas mas.yaml -q "Plan a trip from Celestia to Verdantia"
mas-ctl validate agent.yaml

Full flags: mas-ctl.md. Persist human defaults in config.yaml (mas_ctl.trace: summary).


mas-lab

mas-lab init                          # ~/.config/mas/config.yaml (+ optional infra)
mas-lab config                        # effective paths
mas-lab benchmark run experiment.yaml
mas-lab telemetry show traces/events.jsonl
Command Purpose
mas-lab init User config + optional default LLM infra
mas-lab config Print resolved data / cache paths
mas-lab check / validate Experiment / manifest checks
mas-lab benchmark … Run, list, show, plot, pipeline
mas-lab telemetry … Show / dump / push events.jsonl
mas-lab plot … Trajectory and communication diagrams

Benchmark flag table (repo): lab/docs/cli-reference.md (also mas-lab benchmark run --help). Not published on GitHub Pages.