@eduardogainzakoller/polaris
v0.8.3
Published
AI coding agent for the terminal — works with Codex, Claude and the Anthropic API (Developer Preview)
Maintainers
Readme
Polaris
An AI coding agent for your terminal — works with Codex, Claude and the Anthropic API.
✦ my-project v0.8.3 · Developer Preview
┃ Add validation to createUser and add the tests it needs.
✓ Read src/user/UserService.ts 142 lines · 0.1s
✓ Grep "createUser" 4 matches · 0.2s
✓ Edit src/user/UserService.ts +6 -1 · 0.1s
✓ Write src/user/UserServiceTest.ts created · 38 lines · 0.1s
╭─ Permission required ─────────────────────────────────── enter allow · d deny ─╮
│ Run command │
│ ./gradlew test │
│ Reason: Executes project code. │
╰────────────────────────────────────────────────────────────────────────────────╯Developer Preview. Polaris is early software. It can modify files and run commands in your workspace. Use it inside a Git repository while testing, and review what it changes. Feedback is very welcome — see Reporting problems.
What is Polaris?
Polaris is an interactive command-line coding agent, in the family of Claude Code and Codex. You describe a task in plain language; the agent reads your repository, edits files, runs the commands it needs (with your approval when it matters) and iterates until the work is done.
It runs on top of the AI tools you may already use — the Codex CLI, Claude Code, or an Anthropic API key — and adds a common layer around them:
- Smart permissions — routine reads and the edits your request implies happen without interruptions; Polaris asks only before running project code, installing packages, using the network or destructive Git commands. Nothing outside your project folder is ever touched.
- Safe undo — Polaris tracks exactly what it changed, separately from your own
uncommitted work.
/diffshows its changes,/undoputs them back. - Honest verification — every test or build it runs is recorded; if the code changes after a passing test, the result goes back to unverified.
- Project knowledge — a
POLARIS.mdfile and reusable skills teach it your project's conventions. - Live feedback — you always see what is running, for how long, and its latest output.
Requirements
- Node.js 22 or newer (
node --version). - Git — recommended; Polaris works outside a repository, but
/diffand change tracking work best inside one. - At least one provider (see Providers). The built-in
mockprovider needs nothing and is handy for a first look.
Tested on Windows 11. Linux and macOS are experimental: the test suite runs on them in CI, but they have not had hands-on use yet — reports are especially welcome.
Installation
npm install -g @eduardogainzakoller/polaris@previewThis installs the polaris command. To update, run the same command again; to
uninstall, npm uninstall -g @eduardogainzakoller/polaris.
From source
git clone https://github.com/EduardoGainzaKoller/polaris-cli.git
cd polaris-cli
npm install
npm run build
npm link # makes the `polaris` command available everywhereTo update later: git pull && npm install && npm run build.
To uninstall: npm unlink -g @eduardogainzakoller/polaris.
First run
cd ~/projects/my-app # any project, ideally a Git repository
polarisThe first time, Polaris shows a short notice, checks which providers are available on
your machine, and asks you to pick one. That choice is saved to ~/.polaris/config.json;
you can change it any time with /provider. It never asks for an API key and never stores
one.
Not sure everything is set up? Run:
polaris doctorProviders
| Provider | Uses | You need |
| --- | --- | --- |
| codex | the Codex CLI and your ChatGPT sign-in | the Codex CLI installed; run codex once to sign in |
| claude | the Claude Agent SDK (installed with Polaris) | a Claude Code sign-in (run claude once), or ANTHROPIC_API_KEY |
| anthropic-api | the Anthropic Messages API directly | ANTHROPIC_API_KEY set in your environment |
| mock | nothing — an offline demo that echoes you | nothing |
Choose one for a single run with polaris --provider claude, or switch inside Polaris with
/provider. polaris doctor shows which ones are ready.
Basic usage
Type what you want, the way you would ask a colleague:
❯ Explain how this repository is organised.
❯ Fix the validation bug in UserService.
❯ Add tests for this behaviour and run them.Polaris shows each step as it happens (● Read …, ● Edit …). When it needs your
approval, a card explains why: Enter allows once, d denies. Ctrl+C cancels the
current task without closing Polaris. After a task that changed files, it prints what was
verified.
❯ /diff what Polaris changed
❯ /undo put it back (your own changes are never touched)Permissions
Polaris starts in the smart profile:
| | read-only | smart (default) | workspace-write |
| --- | --- | --- | --- |
| Read, search, git status / git diff | ✓ | ✓ | ✓ |
| Edit and create files in your project | — | ✓ when you asked for changes | ✓ |
| Tests, builds, installs, network, Git commits | — | asks | asks |
| Anything outside your project folder | denied | denied | denied |
"Analyse this service" reads freely but will ask before editing anything; "implement X"
edits without interruptions and stops when it wants to run your test suite. Change
profile with /permissions, or for one run with polaris --permissions read-only.
Project context and skills
POLARIS.mdin your project holds standing instructions — "we use hexagonal architecture", "never edit existing migrations". Polaris reads it at startup.- Skills are reusable procedures in
.polaris/skills/<name>/SKILL.md(per project) or~/.polaris/skills/<name>/SKILL.md(for you). The agent sees only their names and descriptions and loads one when it is relevant;/skill <name>loads one yourself.
The full guide has the file format.
Commands
| Command | What it does |
| --- | --- |
| /help | All commands |
| /status | Version, provider, model, Git and verification state, token usage |
| /provider · /model · /effort | Switch provider, model or reasoning effort |
| /permissions | Show or change the permission profile |
| /tools | What Polaris may do, and when it asks |
| /context | The POLARIS.md instructions in use |
| /skills · /skill <name> | List or load skills |
| /diff | What Polaris changed this session |
| /checkpoint · /checkpoints · /undo | Save and restore Polaris's changes |
| /verify | Run the project's checks against the current changes |
| /activity | What is running right now |
| /new | Start a new conversation (files are kept) |
| /clear | Clear the screen |
| /config | Show or save the configuration |
| /exit | Quit (also Ctrl+C when idle) |
Command-line options: polaris --help.
Troubleshooting
Start with polaris doctor. It checks Node, Git, each provider, your configuration
and the current folder, and says what to do about anything missing. It changes nothing
and never contacts a model.
| Problem | What to do |
| --- | --- |
| Codex CLI not found | Install the Codex CLI, or set POLARIS_CODEX_EXECUTABLE to its path |
| Codex is not authenticated | Run codex and sign in again |
| Claude sign-in has expired | Run claude and sign in again, or set ANTHROPIC_API_KEY |
| Invalid Polaris configuration | Fix or delete the file it names; Polaris never overwrites it |
| Polaris requires Node.js >= 22 | Install a current Node.js LTS release |
| The provider fails to start | Try another one: polaris --provider mock |
| Something looks wrong | Run with polaris --debug and check the log file it names |
Logs live in ~/.polaris/logs/ (the latest 10 sessions). They never contain API keys or
tokens.
Reporting problems
Open an issue at github.com/EduardoGainzaKoller/polaris-cli/issues and include:
- Polaris version (
polaris --version), operating system and Node version - the provider you were using
- the steps to reproduce, what you expected and what happened
- the output of
polaris doctor— or attach the file frompolaris doctor --report, which removes your home directory path and any credentials
Never include API keys or authentication tokens. Polaris sends no telemetry: nothing leaves your machine unless you share it yourself.
Contributing
Setup, tests and architecture are in CONTRIBUTING.md; security issues in SECURITY.md. The full guide documents every behaviour in depth.
