Loading...
Reflag asks you to migrate to a new flag SDK and trusts an LLM to remove the conditionals. FlagShark keeps your existing flag provider and uses deterministic AST transforms. No migration. No hallucinations. Same code, every time.
Reflag is a feature flag platform with AI-assisted removal. To use it you migrate onto their SDK. The cleanup is generated by an LLM, which produces output that varies between runs and occasionally rewrites code in ways that diff reviewers don't expect.
FlagShark is vendor-agnostic. Stay on LaunchDarkly, Split, Unleash, GrowthBook, PostHog, ConfigCat, Flagsmith, or Statsig. We use tree-sitter AST parsing plus deterministic transforms. The same flag always produces the same PR.
You have hundreds, maybe thousands of flag references already in production code. You have client SDKs in mobile apps you can't easily reissue. You have evaluation logic baked into edge functions, server middleware, and feature toggles your team wrote three years ago.
The cost of migrating off your current flag provider to use a cleanup tool dwarfs any cleanup savings. That's why FlagShark works on top of your existing SDK. We don't ask you to rip out anything.
AI-generated cleanup PRs vary run to run. Sometimes they rename a variable that wasn't related to the flag. Sometimes they collapse two unrelated conditionals into a ternary. Sometimes they touch whitespace across the file because the model felt like it.
Every one of those changes is a code review you have to do carefully. The whole point of automated cleanup is that the diff should be obvious: this flag, gone; these lines, removed; this branch, kept.
FlagShark uses tree-sitter to parse your code, identifies the exact AST nodes that reference the flag, and applies a mechanical transform. The output for a given flag never changes. Your team can merge with a 30-second review.
NEW SDK + AI Reflag | ★ DETERMINISTIC FlagShark | |
|---|---|---|
| Works with your existing flag provider | ● | ● |
| SDK migration required | ● | ● |
| Cleanup mechanism | LLM generation | AST transform (tree-sitter) |
| Deterministic output | ● | ● |
| Same input = same PR | ● | ● |
| Languages supported | limited | 12 |
| Flag providers supported | own SDK only | 8+ |
| PR-native review workflow | Dashboard required | Standard GitHub PR |
| Vendor-agnostic | ● | ● |
| Free tier | — | 3 repos, forever |
30-second install. Free for 3 repos. Works on top of every major flag provider.