Skip to main content
ResilientModel wraps any model with automatic key rotation and provider failover. It handles rate limits (429) transparently.

​
Key Rotation

Distribute load across multiple API keys: 
from definable.model.resilience import ResilientModel, KeyPool
from definable.model.openai import OpenAIChat

model = ResilientModel(
    inner=OpenAIChat(id="gpt-4o"),
    key_pool=KeyPool(
        keys=["sk-key1", "sk-key2", "sk-key3"],
        strategy="round_robin",  # or "lru"
    ),
)

agent = Agent(model=model)
When a key gets rate-limited (HTTP 429), it enters cooldown with exponential backoff. The next key in the pool is used automatically.

​
Provider Failover

Fall back to alternative providers when one fails: 
from definable.model.resilience import ResilientModel, FailoverChain, FailoverEntry
from definable.model.openai import OpenAIChat
from definable.model.anthropic import Claude

model = ResilientModel(
    inner=OpenAIChat(id="gpt-4o"),
    failover=FailoverChain(entries=[
        FailoverEntry(model=OpenAIChat(id="gpt-4o-mini"), priority=1),
        FailoverEntry(model=Claude(id="claude-sonnet-4-20250514"), priority=2),
    ]),
)
If the primary model fails, ResilientModel tries each failover entry in priority order.

​
Key Pool Options

​
keys
List[str]
required
API keys to rotate between.
​
strategy
str
default:"round_robin"
Selection strategy: round_robin or lru (least recently used).
​
cooldown_seconds
float
default:"60.0"
Base cooldown time for rate-limited keys.

​
Events

from definable.model.resilience.events import KeyRotatedEvent, ProviderFailoverEvent

model.events.on(KeyRotatedEvent, lambda e: print(f"Rotated to key {e.key_index}"))
model.events.on(ProviderFailoverEvent, lambda e: print(f"Failed over to {e.provider}"))

​
Imports

from definable.model.resilience import ResilientModel, KeyPool, FailoverChain, FailoverEntry