Quickstart

From signup to your first request in 5 minutes.

1. Create an app

1. Sign in at dynoyard.app/login (magic link, no password).
2. Click + New App.
3. Name it (lowercase, hyphens — this becomes your subdomain <name>.dynoyard.app).
4. Pick a blueprint:
   • qwen-coder-32b-awq — coding agents (Zed, Cursor, Continue)
   • qwen-coder-7b — lightweight code completion
   • qwen-7b-instruct — general chat / tool use
   • llama-3.1-8b-instruct — broad chat
5. Pick a tier (see pricing) and click Create.

First boot takes 3–10 minutes (we provision the GPU, install vLLM, download weights). The dashboard shows progress.

2. Send your first request

Once status is ready, copy your API key from the app detail page (shown once on creation — store it like a password).

curl https://my-bot.dynoyard.app/v1/chat/completions \
  -H "Authorization: Bearer sk-dyno-XXXX" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "Qwen/Qwen2.5-Coder-32B-Instruct-AWQ",
    "messages": [{"role": "user", "content": "Write a Python function to reverse a list"}],
    "max_tokens": 256
  }'

3. Wire into your IDE

Zed

Open ~/.config/zed/settings.json:

{
  "language_models": {
    "openai": {
      "api_url": "https://my-bot.dynoyard.app/v1",
      "available_models": [{
        "name": "Qwen/Qwen2.5-Coder-32B-Instruct-AWQ",
        "display_name": "Qwen Coder 32B",
        "max_tokens": 32768
      }]
    }
  }
}

Set OPENAI_API_KEY=sk-dyno-XXXX in your shell. Zed picks it up. Tool-call ready — the blueprint ships with hermes parser pre-configured.

Cursor / Continue

Both accept "OpenAI-compatible" custom endpoints. Use base URL https://my-bot.dynoyard.app/v1 and your sk-dyno-XXXX as the API key.

4. CLI

# Install
curl -fsSL https://pub.dynoyard.app/install.sh | bash

# Login (opens browser)
dyno login

# Create
dyno create my-bot --blueprint qwen-coder-32b-awq --tier pro

# List
dyno apps

# Destroy
dyno destroy my-bot

FAQ

Why is my first request taking forever?

Cold start. We provision a fresh GPU, install vLLM, download model weights (~18 GB for AWQ-32B). Takes 3–10 min for the first request after creation. Subsequent requests on the same instance are sub-second.

What's the difference between Hobby and Standard?

Both run on shared L40S pools. Hobby (20 customers/pool) is for solo devs and small experiments. Standard (12 customers/pool) gives more concurrency headroom — fits a 5–10 person team running an IDE agent in parallel.

What happens when I stop sending requests?

After 10 minutes idle, your app suspends. The next request returns 503 with Retry-After; we resume the slot in the background. Resume is <2 seconds for shared pools (the pool stays warm), ~30 seconds for dedicated. Configure your client to retry once on 503.

Can I bring my own model?

Not yet. Pilot is curated blueprints only. Custom blueprints land in v1.5 — email [email protected] if you have a specific model.

Can I see usage / billing?

App detail page shows last-30-day token usage. Stripe Customer Portal handles invoices (link in dashboard). Billing is monthly per app.

How do I get help?

Pilot customers: email [email protected] — we typically respond within 4 hours during EU/GCC business hours.