Welcome

The Stream CLI is a single binary for working with Stream from your terminal, or for handing that work to your AI agent. It covers every Stream product: Chat, Video, Feeds and Moderation.

The terminal above is interactive: try running getstream commands yourself.

Start here

Quickstart with an agent

Go from install to live data in five minutes, with your agent driving.

Manual quickstart

The same ground, running every command yourself.

How the pieces fit together

You and your agent run the same binary. It authenticates once, calls the Stream API on your behalf, and writes keys and test tokens into your project when you ask it to.

flowchart LR
  you[You] --> cli[getstream CLI]
  agent[Your AI agent] --> cli
  cli <--> api[Stream API]
  cli -. keys and test tokens .-> proj[Your project]

What you can do

Manage apps

getstream init links a directory to one of your apps and getstream open jumps to it in the Dashboard.

Query and act on live data

One command covers around 375 endpoints across every product.

Wire up your project

getstream env writes your SDK keys and getstream token signs test JWTs.

Set up your agent

getstream skills installs the Stream Agent Skills and your agent drives the rest.

Script it

JSON output, --jq and stable exit codes make it safe to pipe and to run in CI.

FAQ

Which products does it cover?

All of them. One getstream api command reaches around 375 endpoints across Chat, Video, Feeds and Moderation; the endpoint catalog lists them.

Does it run on Windows?

There is no native Windows build. Run it in WSL; the install page has the steps.

Do I need Node or Python?

No. The CLI is a single binary.

How does it authenticate?

getstream login opens a browser flow once per machine. CI skips login: set STREAM_API_KEY and STREAM_API_SECRET in the environment.

Is .stream/ safe to commit?

No: .stream/creds.yaml holds the app's secret, so getstream init adds .stream/ to .gitignore for you. See projects and credentials.

Can my AI agent drive it?

Yes, it's built for that. The quickstart goes from install to live data with your agent at the wheel.

Is the output stable enough to script?

Yes. JSON output, --jq filtering and stable exit codes hold up in pipes and CI.

Quickstart with an agent