What is Flagr?

Flagr is an open-source Go service for feature flags, A/B tests, and dynamic configuration. One primitive, the flag, backs all three: your code calls POST /api/v1/evaluation, Flagr looks at who is asking (entityID, entityContext), and returns a variant plus optional JSON attachment.

That lets you decouple deploy from release (ship code dark, turn it on per audience), run experiments with sticky assignment, and change runtime config without redeploying.

openflagr/flagr is the community home of Flagr, continuing development from the original checkr/flagr.

---

Documentation

Site: https://openflagr.github.io/flagr (VitePress)

Contributors: clone the repo and run make help (build, test, UI, CI targets). Docs: make serve-docs (http://127.0.0.1:8081/flagr/) / make build-docs → docs/.vitepress/dist.

Page	Content
Behavioral contracts	Eval vs exposure, recording, eval-only, cache
Integration guide	Eval, batch, exposures (client API)
Contributing	Clone, build, test, OpenAPI
Overview	Concepts, running example, architecture
Use cases	Flags, A/B, dynamic config; GET ?json= eval
Built-in context injection	@ts*, @http_* in entityContext
Self-hosting	Docker, DB, Compose, K8s
Environment variables	DB, auth, recorders (pkg/config/env.go)
Exposure logging	Client impressions for A/B
Data recorders	Kafka, Kinesis, Pub/Sub
API reference	OpenAPI

---

Features

• Feature flags - kill switches, targeted rollouts
• GET evaluation - GET /api/v1/evaluation?json=… (same JSON as POST; use cases)
• Built-in context injection - @ts* and @http_* keys merged server-side (guide)
• Duplicate flag - POST /flags/{id}/duplicate or UI Duplicate Flag
• A/B testing - deterministic assignment; pair with exposure logging
• Dynamic configuration - variantAttachment JSON on eval responses
• GitOps - json_file / json_http; flagr-validate in CI
• Exposure logging - POST /exposures for trustworthy denominators
• Self-hosted - official Docker image + env vars
• Databases - SQLite, MySQL, PostgreSQL, or JSON sources
• Vue 3 UI - TypeScript (browser/flagr-ui); make build-ui, make test-e2e

Quick start

docker pull ghcr.io/openflagr/flagr
docker run -it -p 18000:18000 ghcr.io/openflagr/flagr

open http://localhost:18000

Demo API: try-flagr.onrender.com (may cold-start)

curl -sS -X POST https://try-flagr.onrender.com/api/v1/evaluation \
  -H 'content-type: application/json' \
  -d '{
    "entityID": "127",
    "entityType": "user",
    "entityContext": { "state": "NY" },
    "flagID": 1,
    "enableDebug": true
  }'

Flagr UI

Architecture

Three parts (overview diagram, behavioral contracts):

• Evaluator - POST or GET /evaluation reads EvalCache in memory (default reload 3s; no per-request SQL). Bucketing and stickiness: overview. GET: use cases.
• Manager - CRUD + flag_snapshot rows; webhooks after commit.
• Metrics - async recorders (Kafka, Kinesis, Pub/Sub, Datar); slow sinks do not block eval.

Source: pkg/handler/eval.go, eval_cache.go, crud.go.

Performance

vegeta load test (~2k req/s, sub-ms median in published run):

Requests      [total, rate]            56521, 2000.04
Duration      [total, attack, wait]    28.26s, 28.26s, 365.53µs
Latencies     [mean, 50, 95, 99, max]  371.63µs, 327.99µs, 614.92µs, 1.39ms, 12.50ms
Success       [ratio]                  100.00%
Status Codes  [code:count]             200:56521

Client libraries

Language	Client
Go	goflagr
JavaScript	jsflagr
Python	pyflagr
Ruby	rbflagr

License

• openflagr/flagr - Apache 2.0
• checkr/flagr - Apache 2.0 (original)