dif.sh

What it is

dif.sh is a git-native feature-flag and A/B testing stack for product teams who want experiments to live next to the code they change. We verified www.dif.sh on 2026-09-22 (HTTP 200): each flag or experiment is one Markdown file under a dif/ folder, reviewed in a PR like any other change. dif build emits a typed client your app imports and a context.json that coding agents can read on session start — so prior learnings and active variants travel with the work instead of sitting in a separate dashboard nobody opens.

The same file shape covers plain flags, staged rollouts, holdouts, and split tests. Audience attributes (country, plan, returning visitor, etc.) are declared in config; values arrive at runtime from your app context, so customer lists never get committed. Optional Dif Cloud adds cross-repo pulse and confidence intervals, but proposed conclusions still land as PRs — git stays canonical.

Who it’s for

Indie makers and small product teams already living in PRs and agent-assisted coding who are allergic to “create an account, grab an API key” before a flag exists. If you need multi-language SDKs, instant remote kill-switches without a deploy, or a mature ops dashboard with years of edge-case hardening, a LaunchDarkly-class platform is still the safer default. If your experiments are mostly TypeScript/web and you want CI to fail on colliding exclusion groups, dif’s build-time conflict checks are the interesting part.

Why we listed it

Fresh public launch: Product Hunt day 2026-09-05 (featuredAt 2026-09-05 per PH), live site verified on 2026-09-22. Category gap on this board — we already have billing, email marketing, and plenty of agent tooling, but no git-native feature-flag product. Fits the new-only bar: early open-source CLI + optional Cloud, not a household experimentation brand.

Caveats

We have not run dif through a high-traffic production experiment program. Self-hosted CLI/SDK coverage is JS/TS-forward today; treat other runtimes as “check docs before you bet the roadmap.” Cloud pricing and event quotas on the marketing site are vendor claims — re-read before you depend on hosted lift math. Early-tooling risk: schemas, agent onboarding, and conflict-detection edge cases will still move.