← agnt

Documentation

agnt is an AI coding agent for the terminal. This page covers installing it, connecting a model, and every slash command.

agnt terminal UI in plan mode: it reads files, greps the repo, proposes a rate-limiting change across two files and waits for approval

Install

Linux:

curl -fsSL https://raw.githubusercontent.com/mozzarellabio/agnt/main/install.sh | sh

Windows (PowerShell):

irm https://raw.githubusercontent.com/mozzarellabio/agnt/main/install.ps1 | iex

Connect a model

Hosted providers (bring your own key): Anthropic, OpenAI, OpenRouter, Gemini, DeepSeek, Groq, xAI and Mistral. Local: Ollama and LM Studio.

/key
store an API key
/model
pick a model from any provider
/provider
add an OpenAI-compatible endpoint
/fallback
models to switch to when the current one errors or is rate-limited
/effort
reasoning effort

Modes and permissions

Four modes: ask confirms each action, auto accepts edits, plan reads and proposes until you approve, free has no limits.

/mode
permissions: ask, auto-accept, read-only plan, or free (no limits)
/allow
always allow matching actions, e.g. bash:git *
/deny
always block matching actions, e.g. bash:rm *

Teams of agents

/team
spawn a team of agents (squad, fullstack, research, audit, orchestra)
/spawn
add one agent to the team
/agents
list agents and chat with one
/focus
switch to an agent's chat
/kill
remove an agent
/orch
let the main agent delegate to teammates
/roles
available agent roles
/board
show the shared task board
/map
live map of the agent orchestra

Sessions and context

/new
new chat tab
/close
close this chat tab
/sessions
resume a saved chat
/clear
clear this agent's conversation
/compact
summarize older context now
/context
context window usage
/stats
requests, tokens, cache and cost
/memory
show remembered project facts (.agnt/memory.md)
/export
write this conversation to a markdown file
/copy
copy the last answer to the clipboard
/todo
show the focused agent's task list

Undo and rewind

/undo
restore the working tree to before the last turn (git repos include bash changes)
/redo
reapply what /undo reverted
/rewind
roll files and conversation back to before an earlier prompt
/worktree
work in an isolated git worktree on its own branch
/diff
git diff of the working tree

Git and GitHub

/init
write AGENTS.md for this project
/review
review uncommitted changes (read-only)
/git
branch, upstream, changes and open PR with checks
/branch
create and switch to a new branch
/commit
commit the current changes with a message that matches the repo style
/pr
no number: push and open a PR for this branch; with a number: review that PR (read-only)
/issue
read a GitHub issue, work on it on a branch and verify
/checks
diagnose and fix failing CI on this branch

Extending agnt

/mcp
MCP servers (stdio tools)
/plugin
list, install or remove plugins (commands, agents, skills, skins, hooks, MCP, code mods)
/skills
list available skills (built-in, user, project, plugin)
/schedule
run a prompt later or repeatedly: every 30m, in 2h, daily 09:00, or a date
/loop
repeat a prompt every interval (30s, 5m, 1h) while idle
/rc
control agnt from your phone (QR code)

Look and feel

/skin
change theme
/backdrop
ascii animation behind the transcript; custom frames in ~/.config/agnt/art/*.txt or plugin art/
/caret
toggle the smooth gliding caret
/vim
toggle vim keys in the input
/mouse
toggle mouse capture (off lets you select text)
/sound
toggle a bell when a long task finishes
/settings
show or change settings
/help
commands and keys
/exit
quit