What is Flagr?

Flagr is an open-source Go service for feature flags, A/B testing, and dynamic configuration. One primitive — the flag — covers all three: a decision point in your code that the evaluation engine resolves at runtime based on who is asking.

It exists to decouple deploy from release — turn a feature on for one user, a thousand, or nobody, without redeploying. Run experiments and trust the numbers. Change configuration without a code change or a restart.

openflagr/flagr is the community-driven home of Flagr, advancing development beyond the original checkr/flagr.

---

📖 Documentation

https://openflagr.github.io/flagr

Developers: clone the repo and run make help for build, test, and UI commands (single entrypoint for CI and local work).

Page	Content
Home	Quick start, dev, testing, deploy
Overview	Concepts, running example, architecture
Use Cases	Feature flags, A/B testing, dynamic configuration
Debug Console	UI evaluation testing
Server Configuration	Environment variables, DB, auth, recorders
JSON Flag Source	GitOps, JSON format, validator
Notifications	Webhooks on flag changes
Exposure Logging	POST /exposures API
Data Recorders & A/B Analysis	Kafka, Kinesis, Pub/Sub; sample consumer; A/B analytics
Datar Analytics	In-process eval aggregates
API Reference	Swagger/OpenAPI spec

---

Features

• Feature flags — binary on/off, kill switches, targeted rollouts by audience
• Duplicate flag — clone variants, segments, constraints, distributions, and tags to a new flag (POST /flags/{id}/duplicate or Flag Management on flag detail)
• A/B testing — multi-variant experiments with deterministic, sticky assignment
• Dynamic configuration — per-variant JSON attachments, no redeploy or restart
• GitOps / Flags-as-code — load flags from JSON or HTTP; manage in Git, validate in CI, rollback with git revert
• Exposure logging — POST /exposures for client-reported impressions, the trustworthy A/B denominator
• Webhook notifications — HTTP POST on every flag change, with retry and backoff
• Multi-database — SQLite (dev), MySQL, PostgreSQL, and JSON sources
• Vue 3 UI — TypeScript management UI (Vite, typed REST via ApiResult); see make help for build-ui / test-e2e

Quick start

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

# Open the Flagr UI
open localhost:18000

Or try the hosted demo at https://try-flagr.onrender.com (cold starts may take a moment):

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

Flagr UI

Architecture

Flagr has three components:

• Evaluator — serves evaluation from an in-memory cache of all flags. The cache refreshes periodically (default 3s) and short-circuits when nothing changed, so evaluation never touches the database on the request path.
• Manager — the CRUD gateway; all flag mutations flow through here.
• Metrics — fans evaluation and exposure events out to your data pipeline (Kafka, Kinesis, Pub/Sub) or the built-in Datar aggregates. Recording is asynchronous, so a slow backend never stalls an evaluation.

See the architecture overview for diagrams, request flows, and the deterministic bucketing algorithm.

Performance

Tested with vegeta — 2,000 req/s sustained, sub-millisecond median latency:

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 and Credit

• openflagr/flagr — Apache 2.0
• checkr/flagr — Apache 2.0 (original project)