Retry Settings Configuration

LLxprt Code implements configurable exponential backoff retry logic for API calls to LLM providers. These settings help manage rate limits and transient errors effectively.

Ephemeral settings

Retry configuration uses ephemeral settings, which means they can be changed during a session without modifying your saved profiles or configuration files.

Available retry settings

• retries (number):

  • Description: Maximum number of retry attempts for API calls.
  • Default: 6 for most providers (OpenAI Responses API, Anthropic). Some providers apply their own default — for example, the OpenAI Vercel provider defaults to 2.
  • Example: /set retries 3

• retrywait (number):

  • Description: Initial delay in milliseconds between retry attempts. The delay increases exponentially for subsequent retries.
  • Default: 4000 ms
  • Example: /set retrywait 10000

How to configure retry settings

Use the /set command within the LLxprt Code CLI to configure retry settings for your session:

# Set maximum retry attempts to 3
/set retries 3

# Set initial retry wait time to 10 seconds (10000 ms)
/set retrywait 10000

These settings apply to all subsequent API calls during your session and can be overridden at any time. To persist them, use /profile save.

Provider-specific retry behavior

When you do not set these explicitly, each provider applies its own default. The retry logic includes:

• Special handling for 429 (rate limit) errors, respecting Retry-After headers.
• Automatic detection of transient network issues (socket resets, stream interruptions).
• Integration with streaming pipelines so SSE disconnects are retried without user intervention.