QuickStart - Serverless

This guide takes you from zero to a working OpenAI-compatible endpoint serving a model of your choice.

Prerequisites: a Meshive workspace with credits (QuickStart - Client).

1. Pick a model.

   Open Serverless › Models. The Official tab is a verified, 1-click catalog — each card shows the modality, VRAM requirement, and key specs.

   Want your own model? Register its Hugging Face repo first — Custom models or the full tutorial.

2. Deploy it.

   Click Deploy. The panel lists every GPU candidate that fits, cheapest first, with hourly prices. Four controls:

   • Price cap — max per replica-hour; GPUs above it are never assigned. A suggestion is pre-filled.
   • Min / max replicas — the range (minimum 1).
   • Autoscale — scale within the range by load, or pin to min.
   • Output storage (image/video only) — where generated files land; defaults to Managed.

   Confirm. The deployment appears on Serving as Creating (first deployments show download progress). Running = the endpoint is live.

3. Copy the endpoint URL.

   Every deployment gets its own endpoint, shown at the top of its card on the Serving page:

   https://<your-deployment>.inference.meshive.ai/v1

   This URL is your base_url for everything that follows.

4. Create an API key.

   Go to Serverless › Models → API Keys and click Create. The plaintext key (mk-...) is shown once — copy it now. If you lose it, use Regenerate on the same row to get a new one.

   export MESHIVE_API_KEY="mk-..."

5. Make a request.

   For a chat/LLM deployment:

   curl

   curl https://<your-deployment>.inference.meshive.ai/v1/chat/completions \
     -H "Authorization: Bearer $MESHIVE_API_KEY" \
     -H "Content-Type: application/json" \
     -d '{
       "messages": [{"role": "user", "content": "Hello! Who are you?"}],
       "stream": false
     }'

   Python

   from openai import OpenAI
   
   client = OpenAI(
       base_url="https://<your-deployment>.inference.meshive.ai/v1",
       api_key="mk-...",
   )
   
   resp = client.chat.completions.create(
       model="",  # optional — the endpoint already knows its model
       messages=[{"role": "user", "content": "Hello! Who are you?"}],
   )
   print(resp.choices[0].message.content)

   Node.js

   import OpenAI from "openai";
   
   const client = new OpenAI({
     baseURL: "https://<your-deployment>.inference.meshive.ai/v1",
     apiKey: process.env.MESHIVE_API_KEY,
   });
   
   const resp = await client.chat.completions.create({
     model: "",  // optional — the endpoint already knows its model
     messages: [{ role: "user", content: "Hello! Who are you?" }],
   });
   console.log(resp.choices[0].message.content);

   For an image deployment:

   curl https://<your-deployment>.inference.meshive.ai/v1/images/generations \
     -H "Authorization: Bearer $MESHIVE_API_KEY" \
     -H "Content-Type: application/json" \
     -d '{"model": "<model-id>", "prompt": "A sunset over the ocean", "size": "1024x1024"}'

No code, no setup

Every model also has a Playground (Models page) that runs through the exact same gateway — if it works there, the same request works through the API.

Notes

• model is optional on chat/completion/embedding calls — the endpoint serves one model and fills it in. If set, it must match (GET /v1/models returns the id).
• Billing is GPU-hour × running replicas — pause or delete deployments you’re not using.
• Unknown OpenAI parameters never error — existing OpenAI code ports as-is.

Next steps

• Generate images asynchronously — past the 30s sync budget
• Get a webhook when an async request finishes
• API reference — every endpoint and parameter