Baseten CLI overview

The Baseten CLI manages your Baseten workspace from the command line: models, deployments, environments, secrets, API keys, and raw API access. Every Baseten-native command supports --output json and --jq filtering, so anything you do interactively is scriptable. The CLI is open source on GitHub, where you can browse the code, watch for new releases, and file issues.

The Baseten CLI is in beta and under active development. Commands, flags, output schemas, and the install path may change before general availability.

For authoring Chains, Training jobs, and Loops, use the Truss CLI. The CLI overview explains when to reach for each. For a task-oriented guide to the deployment commands (scale, pause, promote, clean up, logs), see Manage deployments.

Install

Install with Homebrew on macOS or Linux:

brew tap basetenlabs/baseten
brew install baseten

Upgrade later with brew update && brew upgrade baseten.

Optionally run brew trust basetenlabs/baseten before installing to limit how much unofficial tap code Homebrew evaluates by default. See tap trust in the Homebrew docs.

Manual install (Windows, or a specific version)

Download and extract the baseten binary for your platform, then place it on your PATH.

macOS (arm64)
Linux (x64)
Windows (x64)

Download and extract baseten into /usr/local/bin:

curl -sL https://github.com/basetenlabs/baseten-cli/releases/download/v0.3.0/baseten_0.3.0_darwin_arm64.tar.gz \
  | sudo tar xz -C /usr/local/bin baseten

Download and extract baseten into /usr/local/bin:

curl -sL https://github.com/basetenlabs/baseten-cli/releases/download/v0.3.0/baseten_0.3.0_linux_amd64.tar.gz \
  | sudo tar xz -C /usr/local/bin baseten

Download and extract the archive, then move baseten.exe to a directory on your PATH:

Invoke-WebRequest `
  https://github.com/basetenlabs/baseten-cli/releases/download/v0.3.0/baseten_0.3.0_windows_amd64.zip `
  -OutFile baseten.zip; Expand-Archive -Force baseten.zip .

For other platforms or a specific version, download an archive from GitHub releases and move baseten onto your PATH.

Verify the install:

baseten version

Authenticate

baseten auth login

For automation, set BASETEN_API_KEY in the environment instead. See baseten auth for browser login, reading a key from stdin, switching accounts, and credential storage.

Output and filtering

Every Baseten-native command supports four output formats through --output:

text (default): human-readable narrative.
json: a single JSON document. Pair with --jq EXPR to extract one field.
jsonl: one JSON record per line. Used by commands that stream results.
none: suppress stdout entirely.

baseten model push --jq '.predict_url'

--jq implies --output json (or jsonl for streamed commands), so a single flag turns any command into a clean value for the next stage of your pipeline.

Global flags

These flags work on every Baseten-native command:

Flag	Description
`--profile`	Use a specific stored profile for this command, overriding `BASETEN_PROFILE` and the current profile.
`--output`, `-o`	Output format: `text`, `json`, `jsonl`, or `none`. See Output and filtering.
`--jq`, `-q`	Filter JSON output with a jq expression. See Output and filtering.
`--verbose`, `-v`	Enable verbose logging.

Command groups

The baseten CLI organizes commands by resource:

baseten [OPTIONS] COMMAND [SUBCOMMAND] [ARGS]...

Start from the resource you need to manage.

Command group	Use it to
`baseten api`	Make raw management or inference API requests.
`baseten auth`	Log in, log out, switch accounts, and inspect the active account.
`baseten model`	Push, watch, list, describe, predict against, and delete models.
`baseten model deployment`	Activate, configure, download, promote, delete, describe, list, stream logs from, and fetch metrics for deployments.
`baseten model deployment replica`	Terminate an individual deployment replica.
`baseten model environment`	Activate, deactivate, describe, list, stream logs from, and fetch metrics for model environments.
`baseten model-api`	List and inspect Baseten Model APIs.
`baseten org api-key`	List, create, and delete organization API keys.
`baseten org billing`	Inspect organization billing usage.
`baseten org secret`	List, set, and delete organization secrets.
`baseten org team`	List and describe organization teams.
`baseten org user`	List and describe organization users.
`baseten ssh`	Set up SSH access to running Baseten workloads.
`baseten truss`	Forward commands to the `truss` binary on your `PATH`.
`baseten version`	Print Baseten CLI version information.
`baseten whoami`	Show the authenticated user.

Use with AI agents

If you drive the Baseten CLI through a coding agent (Claude Code, Cursor, Codex), add this snippet to your project’s AGENTS.md or CLAUDE.md. The CLI documents itself through --help, so the snippet only covers what an agent can’t discover on its own: how to install, how to authenticate, and how to ask for machine-readable output.

AGENTS.md

# Baseten CLI

The `baseten` CLI manages a Baseten workspace and its models. It never prompts
when stdin isn't a terminal, so it's safe to run headless.

## Install

Check for it first with `baseten version`. If missing, install with Homebrew
on macOS or Linux:

```sh
brew tap basetenlabs/baseten && brew install baseten
```

On other platforms, download the archive for your platform from
https://github.com/basetenlabs/baseten-cli/releases/latest, extract it, and
place `baseten` on your `PATH`.

## Authenticate

To log in, run `baseten auth login --web`: it prints a browser URL for the user
to open and waits for login to complete. For programmatic or CI use, set the
`BASETEN_API_KEY` environment variable instead (create a key at
https://app.baseten.co/settings/api_keys).

## Discover and script commands

- Discover commands with `baseten --help`, then drill down with
  `baseten <command> --help`. Add `--help-output` to see a command's JSON
  output shape and exit codes before parsing the output.
- Script with `--output json` (or `jsonl` for streamed commands) and filter
  with `--jq 'EXPR'`, for example `baseten model list --jq '.models[].id'`.
  `--jq` implies `--output json`.
- Use the `truss` CLI, not `baseten`, to author Chains, Training jobs, and Loops.

The CLI is one piece of the agent toolkit: for the Baseten skill and the MCP servers that let your agent operate a workspace and search these docs, see AI tools.

Next steps

Deploy from CI

Use baseten model push from GitHub Actions and other CI runners.

Compare with Truss

See when to reach for baseten vs truss for each task.

Reference

Inference API

Management API

CLI reference

SDK reference

Training API

Frontier Gateway API

CI/CD

Baseten CLI overview

Install

Authenticate

Output and filtering

Global flags

Command groups

Use with AI agents

Next steps

Deploy from CI

Compare with Truss

​Install

​Authenticate

​Output and filtering

​Global flags

​Command groups

​Use with AI agents

​Next steps

Deploy from CI

Compare with Truss

Install

Authenticate

Output and filtering

Global flags

Command groups

Use with AI agents

Next steps