Slash Commands
UGENT provides a rich set of slash commands for the TUI and REPL. Type / in the input box to see autocomplete suggestions. This guide covers every built-in command, grouped by category.
Quick Reference
Type /help inside UGENT to see all available commands at any time.
Help and System
| Command | Description |
|---|---|
/help (alias /h) | Show all available commands with usage examples |
/version (alias /v) | Show UGENT version information |
/status | Show system status — version, session, tasks, memory, LSP, MCP, plugins, cron, skills. Use /status --full for details |
/reload | Reload workspace configuration and rebuild runtime services. Picks up config file changes without restarting |
/cd <path> | Switch to a different workspace directory at runtime. Requires confirmation — re-enter the command to confirm |
/clear | Clear chat history in the current session |
/quit (aliases /q, /exit) | Exit UGENT |
Provider and Model
| Command | Description |
|---|---|
/setup | Open the setup hub. Subcommands: llm, audio, embedding, context, doctor, legacy |
/providers | Manage LLM instances. list to see all, use <instance> to switch the default instance, model [picker|set <name>] to change its model, model catalog <id>|clear to map it to a catalog entry, policy <instance> user-select|subagent on|off, edit to open config |
/builtin-search | Toggle provider-native built-in web search. on / off, optionally specify a provider instance |
/with <instance[@model]> <prompt> | Send one message to another instance or model. Neither default changes, and the next plain message runs on the default again |
A turn's model is its own. A turn that delegates is resumed on the model it started on once its workers finish, the one /with named included, while that instance is still configured. Channel and web users choose a model of their own with /model, which the TUI and REPL do not have. See Which Model a Turn Runs On.
See also: LLM Providers
Live (Realtime Sessions)
In the TUI, /live holds a realtime session with a voice model. What you say and what it says both appear in the chat, each line once, and the partial transcript shows in the status line. A provider error is shown and the session carries on; it ends when you stop it or the provider closes it. One session is active at a time.
An OpenAI session runs on OpenAI's GA Realtime API, built and read by the same code as a live call in the web workspace. Unlike a call, it has no ask_agent tool and no time limit: you are talking to the voice model alone. /live is a TUI command; the REPL does not have it.
| Command | Description |
|---|---|
/live | Open a session on the default instance |
/live <instance> | Open a session on a named instance (<instance> or <provider>:<instance>) |
/live say <text> | Send a text turn to the active session |
/live stop | End the active session (alias: /live end) |
/live status | Say whether a session is active, and on which instance |
Supported Providers and Models
Realtime models only
The session uses the instance's [media.live] model, or its default_model when that is unset, and the name must contain realtime or live. A chat model is refused as not realtime-capable.
| Provider | Realtime model | Notes |
|---|---|---|
| OpenAI | gpt-realtime-2.1 | Your speech is transcribed with gpt-transcribe; turns are taken by semantic voice activity detection. |
| Azure OpenAI | Your realtime deployment's name | An openai-compatible instance with an Azure base_url; the resource-scoped realtime URL is built for you. api-key or Entra ID auth. |
gemini-3.8-live | Gemini Live. Both sides are transcribed, talking over a reply interrupts it, and a session survives the server's goAway by resuming. Checked against recorded protocol frames only, not yet against the live service. |
Anthropic and Ollama instances cannot hold a session. A session carries audio and text only; no video is sent.
Configuration
[llm.instances.voice]
type = "openai"
api_key = "$OPENAI_API_KEY"
default_model = "gpt-6-luna"
[llm.instances.voice.media.live]
model = "gpt-realtime-2.1" # on Azure, the deployment name
voice = "alloy" # alloy when unset
input_audio = true # capture the microphone (live-audio builds only)
output_audio = true # ask for spoken replies/live reads model, voice, input_audio and output_audio. enabled only decides whether the instance also serves calls in the web workspace, and /live does not read [llm.voice]; see Voice.
Audio Mode
With the live-audio cargo feature, /live captures the microphone (mixed to mono, resampled to 24 kHz PCM16), plays spoken replies, and stops playback when you start talking. Without it, or when no audio device can be opened, /live runs in text mode: send turns with /live say <text> and read the transcript in the chat.
Usage Example
/live voice # Open a realtime session on the voice instance
/live say hello there # Send a text turn
/live status # Check session state
/live stop # End the sessionSee also: Voice for voice messages, read aloud and live calls in the web workspace; OpenAI and Google AI (Gemini) for each provider's realtime settings.
Tools and Integration
MCP Servers
| Command | Description |
|---|---|
/mcp | List configured MCP services and their connection status |
/mcp add <name> stdio <command> [args...] | Add an MCP service with stdio transport |
/mcp add <name> http <url> | Add an MCP service with HTTP transport |
/mcp wizard | Launch interactive MCP setup wizard |
/mcp enable <name> | Enable an MCP service |
/mcp disable <name> | Disable an MCP service |
/mcp test [name] | Test connection to one or all MCP services |
/mcp add http takes a URL and nothing else, so it cannot configure a server that needs a bearer token. For those use ugent mcp add, which accepts --header and --token-ref and works without a REPL — the headless case.
See also: MCP Servers
LSP
| Command | Description |
|---|---|
/lsp | Show LSP server status |
/lsp start | Start a language server |
/lsp stop | Stop LSP server |
/lsp restart | Restart LSP server |
/lsp detect | Detect available language servers for the workspace |
/lsp auto | Toggle automatic LSP startup |
Security
| Command | Description |
|---|---|
/firewall | Show injection firewall status |
/firewall pins [--changed] | Show MCP pin baselines (use --changed to show only changed pins) |
/firewall approve <server> | Approve an MCP server's pin |
/firewall approve --all | Approve all pending MCP pins |
/firewall reset | Reset all MCP pin approvals |
OS Sandbox
| Command | Description |
|---|---|
/sandbox | Show effective sandbox status |
/sandbox status | Effective state and per-class policy summary |
/sandbox profile <class> | Resolved policy for bash, tooling, mcp_server, external_agent, or plugin |
/sandbox test | Prove containment is enforced right now (distinguishes sandbox EPERM from ordinary EACCES) |
/sandbox reload | Re-read sandbox.toml and rebuild |
/sandbox on / /sandbox off | Toggle for this session only |
The sandbox is kernel-enforced containment for every process UGENT spawns. Human-only — never exposed as an LLM tool. See also: OS Sandbox
See also: Security & Firewall
Plugins
| Command | Description |
|---|---|
/plugin | List all plugins and their status |
/plugin enable <name> | Enable a plugin |
/plugin disable <name> | Disable a plugin |
/plugin restart <name> | Restart a plugin worker |
/plugin logs | View recent plugin logs |
Skills
| Command | Description |
|---|---|
/skills | List loaded MCP services and skills |
/skills reload | Reload skills from disk |
/<skill_name> | Execute a loaded skill by name |
See also: Skills
Rules Engine
| Command | Description |
|---|---|
/rules | Show usage for the subcommands |
/rules reload | Re-read the rules file. On a parse error the previous rules stay and the error is only in the log; the reply is always "Rule set reloaded." In the TUI this reloads the copy /rules test uses |
/rules status | Say whether a rules file is configured, and its path. It does not check that the file parsed |
/rules test <message> | Dry-run a message against the loaded rules, as channel cli (tui in the TUI), source type user, with no input_mode |
See also: Rule Engine
Swarm
| Command | Description |
|---|---|
/swarm | Show swarm status |
/swarm init | Initialize swarm node |
/swarm peers | List trusted peers |
/swarm connect <url> | Connect to a peer |
/swarm trust | Manage peer trust |
/swarm task | Dispatch a task to a peer |
/swarm context | Share context with peers |
/swarm revoke | Revoke peer trust |
External Agents
| Command | Description |
|---|---|
/claude | Claude Code session management |
/claude exec [prompt] | Execute a Claude Code session (supports --session, --max-iterations, --yolo, --safe, --inherit, --custom) |
/claude status | Show active Claude Code sessions |
/claude list | List all tracked sessions |
/claude instances | List Claude Code instances |
/codex | OpenAI Codex session management (same subcommands as /claude, plus --full-auto) |
Terminal Multiplexer
| Command | Description |
|---|---|
/herdr | Toggle Herdr terminal multiplexer integration |
/herdr on | Enable Herdr integration |
/herdr off | Disable Herdr integration |
/herdr status | Show Herdr integration status |
Memory and Context
Memory
| Command | Description |
|---|---|
/memory | Show memory system status |
/memory status | Detailed memory store statistics |
/memory remember <text> | Manually record a memory entry |
/memory reindex | Rebuild the memory search index |
/memory compact | Trigger memory compaction |
/memory reembed | Re-embed all memory entries with the current embedding model |
/memory feedback <helpful|not_helpful> | Provide feedback on memory recall quality |
/memory open | Open the memory store directory |
See also: Memory
Context Client
| Command | Description |
|---|---|
/context | Show context client status |
/context status | Detailed context client status (indexed files, sync state, queue) |
/context sync [--since <dur>] | Incremental sync — only changed files. --since accepts durations like 1h, 30m |
/context sync --since-commit <ref> | Sync files changed since a git commit |
/context force-sync [path] | Force full re-sync of all files or a specific path |
/context restart | Restart the context client service |
/context setup | Configure context client settings interactively |
/context enable | Enable context client for this workspace |
/context disable | Disable context client for this workspace |
Session and Checkpoint
| Command | Description |
|---|---|
/new | Start a fresh conversation (archives the current session) |
/resume | Resume a previous session. Shows a picker if no argument, or /resume <number> to pick directly |
/compress (alias /compact) | Manually compress current session context using LLM summarization |
/checkpoints | List all checkpoints for the current workspace |
/restore | Restore a checkpoint. Shows a picker, or /restore [--force] <id-prefix> to restore directly |
/recall | Undo the last user message. If messages are queued, pops the most recent (LIFO). Otherwise aborts any in-flight LLM call and drops that turn from history |
/steering <message> | Add guidance to the currently active LLM turn — injected before the next iteration without interrupting active tool calls |
Scheduling and Automation
Task Management
These act on the session's task list, the plan the agent keeps for its work. Delegated workers are in the Sub-Agents panel (/subagents); see Tasks and Checks.
| Command | Description |
|---|---|
/task, /task status or /task list | Show the task list, numbered, with each task's state |
/task stop <name|index|all> | Stop one or all tasks |
/task continue <name|index|all> | Continue paused, failed or cancelled tasks |
/task force-complete | Mark every task completed, whatever its state |
Cron Jobs
| Command | Description |
|---|---|
/cron | Show cron scheduler status |
/cron list | List all cron jobs |
/cron show <id> | Show details of a specific cron job |
/cron add | Add a new cron job interactively |
/cron add-memory | Add a memory maintenance cron job |
/cron set-schedule | Modify an existing job's schedule |
/cron enable | Enable cron scheduling |
/cron disable | Disable cron scheduling |
/cron task add | Add a task to a cron job |
/cron task edit | Edit a cron task |
/cron task del | Delete a cron task |
/cron del | Delete a cron job |
/cron run <id> | Trigger a cron job immediately |
/cron review | Review recent cron run results and logs |
Loop and Goal
| Command | Description |
|---|---|
/loop <interval> <prompt> | Run a prompt repeatedly at a fixed interval (e.g., /loop 5m check build status) |
/loop <prompt> every <N> <unit> | The same, with the interval after the prompt (e.g., /loop check build status every 10 minutes) |
/loop <prompt> | Self-paced: UGENT schedules each next run and stops when done |
/loop status or /loop list | Show this session's loop and goal runs |
/loop stop <id|all> | Stop one run or all of them; id prefixes are accepted |
/goal <objective> | Set a persistent objective that UGENT works toward across turns |
/goal status | Show the goal, its progress and its latest run |
/goal pause | Pause after the current iteration |
/goal resume | Resume a paused or blocked goal |
/goal clear | Stop and delete the current goal |
Configure verification gates so a goal cannot report completion until your own checks pass. See Autonomous Runs.
Continual Harness
Off by default; set [harness] enabled = true to make these available. Nothing a refinement proposes is applied until you approve it. Over a channel (with allow_channel_review), each user lists and decides only their own proposals, unless global_review_actors lists them for the shared set. See Continual Harness.
| Command | Description |
|---|---|
/refine [focus] | Read this session's trajectory and queue proposed edits to the durable harness. Nothing is applied |
/harness (or /harness list) | Show pending proposals, the approved entries, and recent refinement passes |
/harness show <id> | Full detail for one proposal, including the exact text that would be injected |
/harness approve <id> | Apply a proposal to the entry set. It reaches the model at the next turn |
/harness reject <id> | Discard a proposal without applying it |
/harness rollback <refinement-id> | Undo every approved edit from one refinement pass |
UI Panels
These, /recall and /steering are TUI commands; the REPL does not have them.
| Command | Description |
|---|---|
/monitor | Toggle the monitoring panel (also Ctrl+I) — shows metrics, resource usage, and backpressure |
/debug | Toggle the debug log panel (also Ctrl+D) — includes memory recall and feedback lines |
/status-bar | Switch the bottom status bar widget. Options: mascot, pixel, toggle |
/subagents | Toggle the Sub-Agents panel (also Ctrl+F). Navigate with Up/Down, Enter for detail, Shift+Enter for all, q to close |
/btw | Show read-only status of this session's delegated sub-agents. Out-of-band — never enters the agent's context |