Feature Flags

Ship Safely with Instant Toggles

Gate new models, prompts, and pipelines behind feature flags. Percentage rollouts, user targeting, dynamic configs, and sub-millisecond evaluation -- with unlimited flags on every plan.

feature_flags.pypython
import winnow

# Initialize the SDK (flags are cached locally)
client = winnow.Client(sdk_key="sdk-live-abc123")

# Check a feature gate with targeting
if client.check_gate("new-rag-pipeline", user_id="user_8291"):
    response = new_rag_pipeline(query)
else:
    response = legacy_pipeline(query)

# Read a dynamic config (real-time JSON)
model_config = client.get_config("model-settings")
temperature = model_config.get("temperature", 0.7)
max_tokens = model_config.get("max_tokens", 1024)

# Segment-aware evaluation
segment = client.get_user_segment("user_8291")
# Returns: "enterprise-beta" | "internal" | "general"

Simple Flags, Powerful Targeting

Everything you need to safely roll out AI features, without the complexity tax.

Feature Gates

Boolean gates with percentage rollout, user ID targeting, and custom attribute rules. Roll out to 1% of traffic, then scale up with confidence.

Dynamic Configs

Serve JSON configuration in real time without redeploying. Change model parameters, prompt versions, or pipeline settings instantly across all environments.

User Segments

Define rule-based segments (enterprise, beta, internal) and target flags to specific groups. Combine attribute rules with AND/OR logic.

Sub-Millisecond Evaluation

Flags are evaluated locally from a cached ruleset. No network round-trip on the hot path. Background sync keeps rules fresh without blocking your application.

Flags built for AI workloads

Generic feature flag tools were designed for UI toggles. Winnow flags are built for AI pipelines -- where a single flag might control which model to call, what prompt to use, and how to configure the entire retrieval chain. Dynamic configs let you change any parameter without touching code.

Unlimited flags on all plans, including free
Dynamic JSON configs, not just boolean toggles
Integrated with Experiments for automatic flag-to-experiment promotion
Local evaluation with no latency penalty
Relay proxy support for high-throughput deployments

Deploy fearlessly

Add your first flag in under a minute. Unlimited flags, unlimited environments, free forever.

Start Free