Skip to main content

​
Core Capabilities (Summary)

  • When encountering 429 or upstream exceptions, the platform automatically retries and switches to enabled and healthy channels.
  • Streaming calls provide heartbeat and exception fallback, automatically closing connections on exceptions to avoid hanging.
  • Dynamic rate limiting guardrails peak-shave during high concurrency to protect main routes and task interfaces.

​
How to Use

  • Before use: In the model directory, confirm that the API key can see multiple channels (if not configured, contact an administrator).
  • During calls: Normally use /v1, /claude/v1/messages, /gemini/:version/models/:model, or task interfaces; no additional parameters needed to enjoy automatic retry and switching.
  • Before scaling: Validate with small traffic first; batch or long streaming scenarios can submit in batches to reduce instantaneous concurrency.

​
Request Example (Streaming Chat)

await fetch("https://app.memorylake.ai/v1/chat/completions", {
  method: "POST",
  headers: {
    "Authorization": "Bearer sk-demo123",
    "Content-Type": "application/json"
  },
  body: JSON.stringify({
    model: "gpt-4o-mini",
    messages: [{ role: "user", content: "Please give me an opening script for a product launch event" }],
    stream: true
  })
});

​
Usage Reminders

  • Automatic retry and channel switching depend on enabled and healthy backup channels; switching is unavailable if not enabled.
  • When rate limiting guardrails take effect, speed may be reduced; recommend batching or staggering calls.
  • On streaming exceptions, the platform will close connections; you can supplement retry logic on the client side as needed.