@some-useful-agents/cli

Command-line interface for some-useful-agents. Author, run, schedule, and manage agents from the terminal.

Install

npm install -g @some-useful-agents/cli

Quick start

sua init                         # initialize a project
sua agent run hello              # run an agent (v1 or v2)
sua tool list                    # see available tools
sua examples install             # install bundled examples
sua dashboard start              # open the web dashboard

Commands

• sua agent — list, new, run, chat, webhook, status, logs, cancel, audit, edit, disable/enable, install, reimport
• sua workflow — lower-level agent tools: list, run, replay (--input K=V overrides one input; the rest are the run's own), import, export, status, logs
• sua memory — list, search, pin, forget (what an agent remembers)
• sua usage — what runs cost, by agent and provider
• sua policy — show, check, validate (the tool policy)
• sua behaviors — list, show, validate (Agent Behavior specs)
• sua tool — list, show, validate
• sua examples — install, remove, list
• sua secrets — set, get, list, delete, migrate, check
• sua vars — list, get, set, delete (global variables)
• sua mcp — start, rotate-token, token
• sua schedule — list, validate, status, start
• sua dashboard — start, signin-url
• sua daemon — start, stop, restart, status, logs (dashboard, scheduler, MCP, worker, local model server)
• sua worker — the Temporal worker: start, install-launchagent, launchagent-status, uninstall-launchagent
• sua state — list, du, export, prune (agents' saved state)
• sua outcome — list, show (what runs achieved)
• sua apple — authorize, connect (Apple Foundation Models on macOS)
• sua planner smoke — check the layout planner end to end
• sua init, sua doctor, sua tutorial

What's new in 0.31

• Notebooks work in steps — find → source → check → decide, pausing after each; a find step with no agent of its own uses sua's finder (several narrow searches at once). Save a notebook as a template, share a brief that prints as a PDF, and change its setup on its own Edit page.
• The agents catalog asks what you want to do — /agents answers with keyword matches and sua's picks, shelves the agents you use, and groups the rest by what they do.
• Pick the Codex and Claude model from what each CLI offers your account now (Settings → LLM); sua tells you in the inbox when a model in use is retired.
• Widgets in chat — any agent's reply, and sua's, can include a widget, drawn as it streams, with working buttons.
• Long searches finish — a step out of turns answers from what it read; a loop can run items at the same time (loopConfig.concurrency); a step can keep going if it fails (box on its Edit page).
• Replay with a changed input — sua workflow replay <runId> --from <node> --input K=V reuses the run's other inputs.

See the changelog for everything, and earlier highlights: goal agents, conversations, memory, ask a person, webhooks, cost limits, tool policies, A2UI views, integrations (/tools?tab=integrations), Suggest a layout, MCP servers (/tools?tab=servers), flow control, and 40+ bundled example agents.

See the main repo README and docs/ for full documentation.

License

MIT