docs.
connect okaydesigner
in okaydesigner, press connect an AI, choose okay model and paste your okm_… key. a cloud key needs nothing else. for a model on your own server or Mac, also paste its address (for example http://localhost:8484/v1). okaydesigner sends each job to the best AI on your team; pick the okay model for a job to keep that job private.
the API
OpenAI chat completions. base URL https://www.okaydeploy.com/v1 (your key picks your endpoint) or your endpoint's own URL, https://www.okaydeploy.com/m/<endpoint>/v1. model name okay-model. tools, tool_choice (auto, required or a named function), JSON answers and streaming all work.
| status | means |
|---|---|
| 401 | no key, or a revoked one |
| 402 | no plan, or this month's GPU hours are used up |
| 403 | the endpoint is paused, or the key belongs to another endpoint |
| 503 | the endpoint is waiting to be switched on |
| 502 | the model is waking up: try again in a minute |
install on a Linux server (on-prem)
needs an NVIDIA GPU with 48 GB or more (L40S, A6000, A100, H100), Docker with the NVIDIA container toolkit, and an on-prem licence key from your dashboard.
curl -fsSL https://www.okaydeploy.com/install.sh | sudo sh -s -- okay-license.…
it installs to /opt/okay-model, downloads the model weights once (about 31 GB), starts it with docker compose and prints your server's okm_ key. the model answers on port 8484. to reach it from okaydesigner on the web, serve it over HTTPS (set OKAY_TLS_CERT and OKAY_TLS_KEY in /opt/okay-model/.env) or use the Mac app.
run on a Mac (download)
needs Apple silicon. 32 GB of memory or more runs the full model (about 19 GB on disk); with 16 GB you get the mini model.
curl -fsSL https://www.okaydeploy.com/install-mac.sh | sh -s -- okay-license.…
it installs llama.cpp and Node with Homebrew if they're missing, downloads the model and starts it when you log in. okaydesigner finds it at http://localhost:8484/v1; no key is needed on your own computer. prefer Ollama? the Modelfile sets up the same model; run the gateway in front of it for the licence.
privacy
okaydeploy never stores or logs what you send or what the model answers: only counts (requests, tokens, seconds) for your usage page. each cloud customer gets their own GPU endpoint. on-prem and download models never call home; the licence is checked on your machine.