Tokens-per-minute rate limiting

If you face any problems with proprietary models (like OpenAI models) where you are rate limited, we provide the option to set exponential_backoff=True and to change delay_between_retries to a value in seconds (defaults to 1 second).

For example:

from agno.agent import Agent
from agno.models.openai import OpenAIChat

agent = Agent(
    model=OpenAIChat(id="gpt-5-mini"),
    description="You are an enthusiastic news reporter with a flair for storytelling!",
    markdown=True,
    exponential_backoff=True,
    delay_between_retries=2
)
agent.print_response("Tell me about a breaking news story from New York.", stream=True)

See our models documentation for specific information about rate limiting.

In the case of OpenAI, they have tier based rate limits. See the docs for more information.