Skip to content

CLI & Commands

Startup options

agenticx                          # interactive session (TUI if installed)
agenticx --cli                    # inline REPL with native scrollback
agenticx --tui                    # force the Textual interface
agenticx -c "Summarize this project"   # one prompt, then exit
agenticx --dry-run                # preview loaded configuration
agenticx --protocol anthropic     # startup protocol: openai (default) / anthropic
agenticx --max-turns 20           # override the conversation turn limit
agenticx --version

Subcommands

agenticx app list                 # versioned App Bundles
agenticx app install <path>       # install + permission diff preview
agenticx app info <name>          # publisher, permissions, runtime range, checksum
agenticx app verify <name>        # checksum + runtime compatibility
agenticx app update <name>        # reinstall from the recorded source

agenticx expert                   # list expert templates
agenticx expert download <name>   # merge an expert profile into ./.agenticx/
agenticx expert restore           # undo the most recent merge

agenticx mcp-server               # expose tools over MCP stdio (trusted clients only)
agenticx serve --token <token>    # localhost Client API (default 127.0.0.1:8765)

app install registers the bundle under .agenticx/apps/ but does not activate its profile; expert downloads are role/configuration data, not registered Apps. See Tools, Skills & Experts and Security, Platform & MCP.

Slash commands

Commands are organized in four tiers: Session → Configuration → Capability → Development. /help lists them in this order.

Session

Command Description
/help Show help, grouped by category
/clear Clear conversation history (persists the empty history)
/info Runtime info: provider/model/protocol/base URL/capabilities/privacy
/copy Copy the last reply to the clipboard
/save [path] Save the last reply to a file
/json on\|off Force JSON replies (response_format passthrough)
/compress Replace history with an LLM summary to save tokens
/summarize Summarize the current task into experience memory
/stats Token usage, prompt composition, API/tool counts, duration
/report Trace performance report (calls, success rate, latency)
/feedback good\|bad Mark task satisfaction (experience memory)
/history [subcmd] History stats/list/search/export/clear
/undo Undo the last file write/delete (checkpoint snapshots)
/session [subcmd] Topics: new/list/switch/rename/fork/remove/scene
/privacy [on\|off\|status] Privacy mode
/audit [subcmd] Audit log: recent/query/clear
/voice [subcmd] Voice input: on/off/model/silence/max/threshold
/tts [subcmd] Voice output: on/off/model/voice/speed/say/last
/camera capture\|record Camera photo / video (media extras)
/log [subcmd] Full prompts recorded under /debug on: tail/search/clear
/trace [subcmd] Execution trace: tail/search/clear
/exit Exit (aliases /quit, /q)

/v is the voice-input shortcut (Ctrl+T): record → ASR → send.

Configuration

Command Description
/setup Interactive workspace initialization wizard
/provider Wizard: provider → model → connection → API key
/model [name] Switch model within the current provider
/capability [subcmd] list/enable/disable/reload/config
/subagents [subcmd] list/reload/enable/disable/config
/trust [subcmd] This turn's tool trust cache: list/clear/allow/deny
/rule [subcmd] Long-lived operation rules: list/add/remove/edit/enable/disable/clear
/theme [subcmd] list/set/custom colors (dark/light/monokai/dracula/nord/solarized)
/debug [on\|off\|status] Record full sanitized prompts to the session trace
/directives [subcmd] Custom directives + auto-learning proposals/accept/reject

Capability

Command Description
/profile [subcmd] User profile: list/search/add/remove/clear
/memory [subcmd] Conversation history memory: list/search/remove/clear
/mcp [subcmd] MCP services: add/remove/list (requires bailian.mcp)
/skill [name] [args] List or invoke skills
/skill install\|link\|update\|uninstall Skill package lifecycle
/skill search\|publish\|reload Registry search, packaging, rescan
/skill enable\|disable <name> Force skill package state
/app list\|info\|install\|update\|verify Versioned App Bundle management
/cron [subcmd] Scheduled tasks: add/list/remove/pause/resume

Commands for disabled capabilities are intercepted with an enable hint instead of failing on credentials.

Development

Command Description
/dev model add\|rm\|list Register models at runtime
/dev provider add\|list\|rm Extend providers
/dev capability add\|list\|rm Extend HTTP tool groups
/dev skill add\|list\|rm Extend skills
/dev tool add\|list\|rm Extend shell tools
/dev debug tools\|schema\|call\|prompt Debugging helpers
/dev test provider <name> Test provider connectivity
/dev reload Hot-reload custom-extensions.toml
/dev log Tool registration statistics
/expert [list] List expert templates
/expert download <name> Merge an expert into ./.agenticx/ (backup on conflict)
/expert restore Undo the most recent merge

Terminal output

  • Markdown replies with numbered turns; tool details collapse by default (Ctrl+O toggles), failures stay expanded.
  • Ctrl+G expands token/API/tool/prompt-composition statistics. Prompt sizes are characters, not exact token counts.
  • Selection copying: Cmd+C on macOS, Ctrl+C on Windows/Linux; on macOS Ctrl+C clears/cancels/quits instead. /copy copies the latest reply source; --cli gives native scrollback.