// reference

CLI reference

Run Astra interactively in a TUI, or headless with -p for scripts, hooks, and CI.

Flags

npm run dev                                    # Interactive TUI
npm run dev -- -p "List files in src"          # Print mode (no TUI)
npm run dev -- --boss                          # Start in boss orchestrator mode
npm run dev -- --boss -p "refactor auth module" # Boss mode, print and exit
npm run dev -- --multi                         # Start in multi-agent (parallel) mode
npm run dev -- --multi -p "add tests for utils" # Multi-agent mode, print and exit
npm run dev -- -c                              # Continue last session
npm run dev -- --model groq/llama-3.3-70b-versatile "hello"
npm run build && npm start                     # Production build
FlagDescription
-p, --printPrint the response and exit — no TUI.
-c, --continueResume the most recent session.
--model <ref>Start with a specific provider/model-id.
--bossEnable the boss orchestrator mode.
--multiEnable multi-agent (parallel) orchestrator mode.
--no-compactDisable context compaction.

Interactive commands

Inside the TUI, prefix any of these with /:

CommandDescription
/model, /mOpen the model selector, grouped by provider.
/buildSwitch to Build mode (full tool access).
/planSwitch to Plan mode (read-only exploration).
/bossToggle the boss orchestrator.
/multiToggle the multi-agent (parallel) orchestrator.
/agentsList agent runs from the current multi-agent session.
/tasksShow the active task plan.
/compactSummarize older messages to free context.
/rulesList loaded project rule files (AGENTS.md, etc.).
/permissionsShow merged permission presets for this project.
/traceShow the path to the latest worker trace log.
/sessionsBrowse and load saved chat sessions.
/settingsThinking level and API key status.
/connectConfigure a gateway connection (Telegram, Discord, WhatsApp).
/skillsBrowse and install skills.
/skill <name>Load a skill for the current turn.
/voiceSpeak a task (voice input).
/newStart a new session.
/quitExit.

Headless commands

Twitter batch runner

Post and reply on X/Twitter fully unattended — zero approvals — from a JSON schedule:

astra twitter-batch schedule.json

See Tools → Twitter for the schedule format and report output.

Testing & evaluation

npm test                 # Run the test suite
npm run test:watch       # Watch mode
npm run test:coverage    # Coverage report
npm run eval             # Agent evaluation suite (smoke, optional)
npm run eval:full        # Full agent evaluations (expensive)
npm run eval:deterministic  # Deterministic evals (no API keys)