Skip to main content

Resilience

Agentix includes configurable retry logic and circuit breakers to handle transient LLM failures and multi-worker deployments.

Retry configuration​

from agentix import AgentixAgentOptions

options = AgentixAgentOptions(
    retry_max_attempts=3,     # default: 3
    retry_base_delay=1.0,     # initial delay in seconds (doubles each attempt)
    retry_max_delay=30.0,     # max delay cap
)

Circuit breaker​

The circuit breaker prevents cascading failures by stopping LLM calls once a threshold of consecutive failures is reached.

options = AgentixAgentOptions(
    circuit_breaker_failure_threshold=5,  # default: 5; 0 = disabled
    circuit_breaker_cooldown=60.0,        # seconds before a probe is allowed
)

When the circuit is open, AgentixResilienceError is raised.

Circuit breaker backends​

By default each AgentixClient instance uses an in-memory circuit breaker (no shared state). For multi-worker deployments, use a shared Redis backend:

from agentix import AgentixClient, AgentixAgentOptions
from agentix.resilience import (
    InMemoryCircuitBreakerBackend,
    RedisCircuitBreakerBackend,
    AsyncRedisCircuitBreakerBackend,
)

# In-memory (default, per-instance)
client = AgentixClient(
    options=options,
    circuit_breaker_backend=InMemoryCircuitBreakerBackend(),
)

# Redis (sync) — shared across workers
client = AgentixClient(
    options=options,
    circuit_breaker_backend=RedisCircuitBreakerBackend(url="redis://localhost:6379/0"),
)

# Redis (async) — uses redis.asyncio
client = AgentixClient(
    options=options,
    circuit_breaker_backend=AsyncRedisCircuitBreakerBackend(url="redis://localhost:6379/0"),
)

Multi-tenant circuit breakers​

The circuit breaker key is f"{tenant_id}:{options.provider}". Use tenant_id to scope state across tenants so one tenant's failures cannot trip another's circuit:

client = AgentixClient(
    options=options,
    tenant_id="customer-abc",
    circuit_breaker_backend=shared_redis_backend,
)

Fallback model​

Configure a fallback model for when the primary model fails after all retries:

options = AgentixAgentOptions(
    provider="anthropic",
    model="claude-opus-4-20250514",
    fallback_model="claude-sonnet-4-20250514",
)

Token budget​

Set a hard stop when total token usage exceeds a budget:

options = AgentixAgentOptions(
    max_tokens_budget=100000,  # hard stop at 100k total tokens
)