Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.deck.co/llms.txt

Use this file to discover all available pages before exploring further.

Deck enforces two types of limits: rate limits on API requests and concurrency limits on sessions running in agent sandboxes.

​
Limits

The rate limit is 200 requests per 10 seconds per organization. Limits are applied across all API keys in your organization. If you have multiple services making requests with different keys, they share the same quota.

​
Rate limited responses

When a request is rate limited, Deck returns: 
{
  "errors": [
    {
      "type": "rate_limit",
      "code": "rate_limit_exceeded",
      "message": "Too many requests."
    }
  ],
  "request_id": "req_a1b2c3d4..."
}

​
Handling rate limits

Implement retries with exponential backoff when you receive a 429 response. 
async function requestWithBackoff(url, options, retries = 5) {
  for (let attempt = 0; attempt < retries; attempt++) {
    const res = await fetch(url, options)

    if (res.status !== 429) return res

    const delay = Math.min(1000 * Math.pow(2, attempt), 30000)
    await new Promise(r => setTimeout(r, delay))
  }

  throw new Error('Rate limit retries exhausted')
}

​
Session concurrency

Each task that Deck runs executes in an agent sandbox tied to a session. The number of sessions your organization can run concurrently depends on your plan. This limit applies across all sessions in your organization regardless of connector or credential. When you submit a task that would exceed your concurrency limit, Deck returns a 429 status code: 
{
  "errors": [
    {
      "type": "rate_limit",
      "code": "session_limit_exceeded",
      "message": "Agent sandbox limit exceeded."
    }
  ],
  "request_id": "req_a1b2c3d4..."
}
Retry with backoff until a running session reaches a terminal state and a slot becomes available.

​
Best practices

  • Use events instead of polling. Register an event destination and let Deck push status updates to you rather than polling GET endpoints in a loop.
  • Paginate list requests. Use limit and cursor to fetch results in pages rather than requesting large datasets in a single call.
  • Stagger task submissions. If you have many tasks to run, queue them and submit the next one after the previous completes. Use events to know when a session finishes rather than polling.