flagd

flagd is a feature flag daemon from the Cloud Native Computing Foundation's OpenFeature project, built around a Unix-philosophy single-purpose process. The maintainers describe it as a ready-made, open-source, OpenFeature-compliant feature flag backend: it loads flag definitions from one or more sources, evaluates them in near real time, and serves results over HTTP or gRPC. Because it implements the OpenFeature contract, applications can use the OpenFeature SDK for their language and swap flag sources or vendors without rewriting flag-checking code.

Flags are defined in JSON against a versioned schema with state, variants, defaultVariant, targeting rules, and metadata. Targeting conditions are evaluated against the request context and support deterministic fractional rollouts, semantic version comparison, and string comparison, so percentage releases and audience targeting work without a separate experimentation service. flagd can merge definitions from several sources at once: local files watched for changes, HTTP endpoints polled with ETag support, gRPC sync streams, Kubernetes FeatureFlag custom resources, and GCS, Azure Blob, or S3 buckets. Changes to a watched file or endpoint are picked up without restarting the daemon.

flagd is distributed as a static binary for x86 and ARM, an OCI image at ghcr.io/open-feature/flagd, a systemd unit, and Kubernetes manifests for sidecar or central-service deployments. The companion OpenFeature Operator handles deeper Kubernetes integration, injecting flagd and flag configuration into workloads. Telemetry is built in through OpenTelemetry metrics and traces, with Prometheus scraping available on the management port. There is no admin UI, database, or hosted tier: flags live as JSON in git, CI pipelines, or object storage. Compared with LaunchDarkly, Split, ConfigCat, Optimizely, Statsig, and Firebase Remote Config, flagd removes the vendor from the evaluation path entirely and asks teams to bring their own authoring and approval workflow.