RemiDocs

Remi CLI

Remi in your terminal, on macOS, Windows and Linux. Ask Remi anything, and let it work on the code in the folder you're in.

Coming soon

The first Remi CLI release isn’t published yet, so the install command has nothing to download today. These are the steps for when it is.

Run remi in a project folder and ask for what you need. Remi reads the code, makes the change, runs the checks and tells you what it did. It asks before it changes anything, unless you tell it not to, and you can undo what it changed.

It's the same Remi as on the web, your phone and Remi Desktop. Your account, models, chats, inbox, workflows and connected apps come with you.

Before you start

  • macOS, on Apple silicon or Intel.
  • Windows, on x64 or Arm64.
  • Linux, on x64 or Arm64, including Alpine and other musl-based systems.
  • A Remi account. Sign up on the web if you don't have one.

You don't need admin rights. Remi installs into .remi/bin in your home folder.

Install

Open Terminal and paste:

curl -fsSL https://remiapp.ai/install.sh | sh

Check the version

remi --version

You should see a version number, such as 0.1.0. If your terminal says it can't find remi, open a new terminal window: the installer adds Remi to your PATH, and only new windows pick that up.

Sign in

remi login

Your browser opens remiapp.ai/device. Check that the code matches the one in your terminal, then select Approve. Over SSH, remi login shows a code instead: open remiapp.ai/device on any device and enter it there.

To sign in with a code sent to your email instead, run remi login --email you@example.com.

Start Remi

Go to a project folder and start Remi:

cd my-project
remi

The first time you run Remi in a folder, it asks whether you trust the folder. Say yes for your own projects. Trust a folder explains what that allows.

Then ask a question, or ask for a change. Before Remi edits a file or runs a command, it shows you what it wants to do and waits for your yes.

Check it worked

remi whoami

This shows the email you're signed in with. For a fuller check, run remi doctor. It checks everything Remi depends on and tells you what to do about anything that's wrong.

Work in a folder

Type your request and press Enter. Remi can read, search, edit and create files in the folder, and run commands there. Each change shows as it happens.

  • Answer a request. When Remi wants to edit a file or run a command, it shows what and waits. Choose Yes; or yes and don't ask again for that command in this folder (for an edit, accept edits for the rest of the session); or No, and tell Remi what to do instead.
  • Undo. /undo puts back the files Remi changed in its last turn.
  • Stop. Press Esc to stop Remi mid-answer. Press Ctrl+C twice to quit.
  • Send a file. Type @ and a path, such as @src/app.ts, to send that file with your message. To send a picture, paste it with Ctrl+V.
  • Run something yourself. Start a line with !, such as ! npm test. Remi sees the output with your next message.
  • Leave a note for next time. Start a line with #, such as # always use pnpm. It's added to the folder's AGENTS.md, which Remi reads in every session.
  • Write more than one line. Press \ then Enter, or Alt+Enter.

Choose how much Remi does without asking

Press Shift+Tab to move between the modes. The current mode shows under the input box.

ModeWhat Remi does without asking
DefaultReads and searches. It asks before every edit and command.
Accept editsEdits files. It still asks before commands.
PlanOnly reads, and answers with a plan. Nothing changes.
AutopilotEdits files, and runs commands in a sandbox that can only write in the project and can't reach the network. It still asks before anything the sandbox can't contain, such as installing packages or touching files that may hold secrets.

On Windows, and on Linux without bwrap installed, there's no sandbox, so Autopilot asks before commands. To start in a mode, use remi --mode plan (or accept_edits, autopilot). Permissions has the details, and how to allow particular commands.

Commands in the app

Type / to see them all. The ones you'll use most:

CommandWhat it does
/modelPick the model, from the models on your account.
/modePick a mode. /autopilot switches straight to Autopilot.
/undoPut back the files Remi changed in its last turn.
/clearStart a new chat.
/resumePick up an earlier session in this folder.
/contextSee what goes with each message: instructions, skills, tools, and how full the conversation is.
/memoryList the instruction files. /memory edit opens the folder's AGENTS.md in your editor.
/answerGo through a form Remi asked you, one question at a time.
/handoffOpen this chat on the web or your phone.
/permissionsSee and change what runs without asking.
/mcpSee your MCP servers and their tools.
/pluginSee the plugins loaded here.
/inboxYour Remi inbox, as the bell shows it on every device.
/workflowsYour workflows, and what they did lately.
/saveSave an image, video or audio Remi just made into the folder.
/copyCopy Remi's last answer.
/exportCopy the whole chat as Markdown, or save it to a file.
/usageHow much of your plan's allowance you've used.
/initWrite a first AGENTS.md for the project.
/reviewReview your uncommitted changes.
/helpEvery command and key.

To pick up where you left off, run remi --continue for the last session in this folder, or remi --resume to choose one.

Ask once, from a script

remi -p answers once and exits, for scripts and automation:

remi -p "Summarise what changed since the last release"
  • For a program to read: add --output-format json for one JSON result, or --output-format stream-json for a stream of events.
  • To let it act without asking: add --allow-tools with the commands it may run, such as --allow-tools "Bash(npm test:*),Edit".
  • To keep it in bounds: --max-turns 5 stops after five rounds of tool calls.

To run Remi where nobody can sign in, such as CI, create a token and set it as REMI_TOKEN:

remi token create --name ci

Scripts and CI has the rest.

Pick up anywhere

While remi is open in a folder, you can carry on that chat from your phone, the web or another computer. What you send from there runs here, in this folder, with this folder's rules. When Remi needs your permission, you can answer on your phone, but the terminal has the final say.

  • /handoff opens the chat on the web or your phone.
  • /host off stops taking messages from elsewhere for this session. remi --no-host starts without it.

Your Remi account in the terminal

CommandWhat it shows
remi inboxWhat came into your Remi inbox. Reading it here marks it read everywhere.
remi workflowsYour workflows and what ran lately. remi workflows run <name> runs one, and pause or resume work the same way.
remi studioThe designs and writing you made in Remi's studio. remi studio export <name> --pdf saves one into this folder (also --png, --pptx, --docx or --md).
remi modelsThe models you can use.
remi usageHow much of your plan's allowance you've used.
remi mcp add <url>Adds a hosted MCP server to your account, for every Remi app. MCP servers has more.

Use Remi from your editor

Zed, JetBrains IDEs, Neovim and Emacs can run Remi as their AI agent, through remi acp. Your editor shows the chat, the changes and the permission requests, and Remi works in the project the editor has open. Connect your editor has the setup for each.

Update or remove

To update, run remi update. Remi also checks for a new version once a day and tells you when there is one.

To remove it, sign out with remi logout, then delete the program:

rm ~/.remi/bin/remi

Your settings are in the .remi folder in your home folder; delete it to remove them too. Your chats stay in your Remi account.

On this page