v0.3.2 — now on npm

Arti

A terminal coding agent powered by Arti AI.
Read files, propose edits, run commands. All from your terminal.

$ npm install -g arti-cli
or
$ npx arti-cli
see it in action

Your terminal, amplified

Try it out. Type a question or command and see how Arti responds.

arti — ~/my-project
arti arti-0.5 ~/my-project [email protected]
type / for commands · @ for files · ctrl+c twice to exit
>

Built for coding

File intelligence
Reads and understands your source files. Explains code, spots bugs, and answers questions about your codebase.
Safe edits
Shows a diff before applying any change. You approve every edit. Nothing touches your filesystem without your say-so.
Command runner
Runs shell commands with your approval and feeds the output back to the model, so it can iterate on test failures or build errors.
One-shot mode
Skip the session and pipe a task directly: arti "summarise this repo". Great for scripts and quick queries.
Ask and edit modes
Switch between ask mode for questions and edit mode for making changes. Toggle with Shift+Tab at any time.
Secure auth
Tokens refresh automatically every hour. Credentials cached in ~/.arti-code/, readable only by you.

Up and running in minutes

Node 20+ required. Everything else is handled automatically.

1
Install the CLI
Install globally with npm. This puts the arti command on your PATH.
bash
npm install -g arti-cli
Or try without installing:
bash
npx arti-cli
2
Log in
On first launch, Arti prompts for your Arti account credentials. Everything is cached so you only do this once.
terminal
$ arti Arti — Login ? Email: › [email protected] ? Password: › •••••••• ✓ Signed in as [email protected]
3
Start coding
Navigate to your project and run arti. Type any question at the prompt, or use @filename to give it file context.
terminal
$ cd my-project $ arti read package.json and tell me what this project does find all TODO comments across src/ there's a bug in line 42 of utils/parser.ts, fix it run npm test and fix any failures
Or run a one-shot task without opening a session:
bash
arti "summarise this repo in one paragraph" arti "what tests are failing?" --model arti-1
4
Review and approve
Whenever Arti wants to edit a file or run a command, it shows you exactly what it plans to do and waits. Type y to confirm or n to skip.
terminal
# Edit proposal - const match = str.match(/\"[^\"]*\"/g); + const match = str.match(/\"(?:[^\"\\]|\\.)*\"/g); Apply this edit to utils/parser.ts? (y/n) y utils/parser.ts updated

Session commands

Type these at the prompt inside an Arti session.

Command Description
/helpShow all available commands
/modeSwitch between ask and edit mode
/btw <question>Ask a quick question without polluting your main context
/compactSummarize history to free up context window
/initCreate .arti/arti.md project memory file
/skillsList skills loaded from .arti/skills/
/clearClear conversation history, keep the session open
/usageShow token usage for this session
/exit  /  /quitExit Arti

Tools Arti can use

Tool What it does Approval required
read_fileRead a file's full contentsNo
list_dirList files and directoriesNo
edit_fileShow a diff and apply a targeted editYes
create_filePreview and create a new fileYes
run_commandShow the command, run it, return outputYes

Pick your power level

Switch mid-session with /model arti-1 or set a default for one-shot runs.

arti-0.3
fast
Lightweight and snappy. Best for simple lookups, quick edits, and tasks where speed matters more than depth.
arti-0.5
★ default
Balanced performance. Handles most coding tasks well. Right choice if you're unsure which model to use.
arti-1
most capable
Maximum reasoning. Best for complex refactors, debugging tricky issues, and multi-file architecture decisions.
bash
# Set model for a single run arti --model arti-1 "refactor this auth module" # Switch model mid-session /model arti-1