Bring Octer models into your agent and your code

Everything below — the agent, images, video, direct API calls — uses this one key.

• Open the Octer console: octer.ai/workspace/o/ → (register first if you don't have an account)
• On the OClaw page, click copy to get a key starting with sk-.

🎁 New users get trial credits on sign-up — grab your key and try everything for free before spending anything. Sign up & claim credits →

Take a second to check the key works (pick your OS first, then copy — a JSON response means both the key and your network are fine):

Replace sk-YOUR_KEY in every command below with your own key.

No JSON back, or a 401 / 403? "Troubleshooting →" has the fix for each case.

OpenClaw and Hermes are free, open-source local AI assistants — they run on your own computer and you chat with them in a terminal. Never installed one? Step 1 below walks you through it from scratch.

Pick your agent and your OS — every command below switches to match.

Step 1 · Install the client

Check the CLI is available:

Step 2 · Switch it to an Octer model

The script writes the endpoint, key and model, restarts the gateway and sends a test “hello” — a normal reply means you're done.

Step 3 · Verify

Run these two, one after the other:

Changed your mind? Undo it

This clears the Octer model config and leaves the rest of your setup untouched.

The step above only covers chat. To let your agent make images and video too, add the oclaw skill — a small CLI that wraps all three, saving what it generates straight to your machine.

The easy way: paste this into your OpenClaw / Hermes chat and the agent installs it itself.

Don't have OpenClaw / Hermes yet? See "Connect your agent →" — two commands and you're set. Prefer no agent? Use the terminal method below.

Swap in your own key. After that, just say “make me an image of…” or “generate a video of…” in plain language — the agent takes it from there.

You can also drive it from the terminal (no dependencies beyond Python 3):

On Windows: this tool is a bash script — run it inside WSL or Git Bash (it won't run directly in PowerShell).

• Images land in images/, videos in videos/.
• Video is asynchronous: it usually takes 1–3 minutes, and the CLI waits and downloads for you. If it times out, nothing is lost — the job is still running. Run ./oclaw.sh status to find the task id, then ./oclaw.sh watch <task-id>.
• Video links expire in about 24 hours, so download as soon as the job finishes. The CLI already does this — if you call the API yourself, don't just store the URL.

No agent needed — to wire Octer into your own program, just call the HTTP API. The gateway speaks the OpenAI protocol, so an official SDK works with nothing more than a new base_url.

The model id is what goes into an API call or an agent config — it's the highlighted string in the table; click the button next to it to copy.

Swapping between models of the same type is just a different model id. Chat and reasoning share one endpoint; images and video have their own — the snippets above switch automatically with the model you pick.

Any client that speaks the OpenAI protocol can reach Octer with these four values.

The two paths you didn't pick live here — the same key works for all of them; expand one whenever you need it.

About your data: prompts and reference images are sent to the Octer gateway to generate the result; your API key only ever travels in the request header and is never written to disk; generated images and video stay on your machine.