What Claude (Anthropic) with web search answered for feature flag platforms (agencies) on 2026-09-24

Claude (Anthropic) with web search claude-opus-5 asked 2026-09-24

Coverage: Claude (Anthropic) with web search. Other engines not included.

The question asked, verbatim: best feature flag platforms for agencies

Brands named, in the order the engine gave

1. ConfigCat recommended
   ConfigCat - my default recommendation for agencies.
2. Flagsmith mentioned
   Flagsmith - the best pick when you need to hand the instance to the client at the end of the project.
3. Unleash compared
   (flagsmith.com/open-source (https://www.flagsmith.com/open-source), Flagsmith vs Unleash (https://www.flagsmith.com/compare/flagsmith-vs-unleash))
4. PostHog mentioned
   PostHog - worth naming because agencies rarely need only flags.
5. GrowthBook mentioned
   GrowthBook - pick this when the client engagement is actually about experimentation and conversion work, not just safe rollouts.
6. LaunchDarkly mentioned
   LaunchDarkly I'd name mainly to tell you when not to use it.

Sources the engine cited

The answer, as recorded

For an agency the ranking is different than for a product team - what matters is many separate client projects, rotating contractors on the dashboard, and being able to hand the whole thing over when the engagement ends. That pushes against per-seat and per-MAU pricing. **1. ConfigCat** - my default recommendation for agencies. Unlimited seats on every paid plan and unlimited projects, so adding a client's PM or a freelance dev costs nothing, and each client gets its own project with isolated SDK keys per environment. Pricing is by config fetches, which scales with actual usage rather than headcount. The catch: the free tier caps you at two environments, so you'll be on a paid plan quickly with real clients. ([configcat.com](https://configcat.com/), [ConfigCat multi-tenant targeting](https://configcat.com/blog/how-to-target-features-by-tenants/)) **2. Flagsmith** - the best pick when you need to hand the instance to the client at the end of the project. Core flagging, targeting, multivariate and remote config are BSD-3-Clause and self-hostable with no user restrictions, so you can run it on the client's own infra and walk away without leaving them a subscription they didn't ask for. Also the right answer for clients with data-residency or on-prem requirements. Free cloud tier is single-user, so cloud is really only for solo work. ([flagsmith.com/open-source](https://www.flagsmith.com/open-source), [Flagsmith vs Unleash](https://www.flagsmith.com/compare/flagsmith-vs-unleash)) **3. Unleash** - the other strong self-host option, and stronger than Flagsmith if a client has genuinely high traffic; it's API-first with a rich activation-strategy model. Caveat for agencies: there's no free SaaS tier and Enterprise is ~$75/seat/month with a 5-seat minimum, which is exactly the shape that hurts when you're staffing a rotating cast onto a dashboard. Use the self-hosted OSS version. ([getunleash.io/pricing](https://www.getunleash.io/pricing)) **4. PostHog** - worth naming because agencies rarely need *only* flags. If you're already standing up analytics or session replay for the client, flags come bundled and it's one less vendor on the invoice. Weaker on release governance (approvals, audit trails) than the dedicated tools. ([posthog.com open-source flag tools](https://posthog.com/blog/best-open-source-feature-flag-tools)) **5. GrowthBook** - pick this when the client engagement is actually about experimentation and conversion work, not just safe rollouts. Open source and self-hostable; the Pro cloud plan is $199/month for up to 20 seats, which is a reasonable flat-ish number if you keep clients in one org. ([growthbook.io](https://www.growthbook.io/blog/best-open-source-feature-flagging-tools-compared)) **LaunchDarkly** I'd name mainly to tell you when *not* to use it. It's the best-governed platform in the category and the right call if a specific enterprise client mandates it - but it's MAU-priced, contracts typically start around $25K/year, and a 50K-MAU client can land in the $100K-150K/year range. That's a client-budget line item, not an agency tool. If a client insists, put the contract in the client's name, not yours. ([costbench pricing benchmarks](https://costbench.com/stats/feature-flags-pricing-2026/), [Rollgate pricing comparison](https://rollgate.io/blog/feature-flags-pricing-comparison)) One practical note independent of vendor: whichever you choose, create a separate project per client rather than using environments or tenant attributes to separate them. Environment-level separation makes a mis-scoped targeting rule capable of leaking one client's flag state into another's, and it makes handover messy. Sources: - [ConfigCat](https://configcat.com/) - [ConfigCat - multi-tenant feature targeting](https://configcat.com/blog/how-to-target-features-by-tenants/) - [Flagsmith - open source](https://www.flagsmith.com/open-source) - [Flagsmith vs Unleash comparison](https://www.flagsmith.com/compare/flagsmith-vs-unleash) - [Unleash pricing](https://www.getunleash.io/pricing) - [PostHog - best open-source feature flag services](https://posthog.com/blog/best-open-source-feature-flag-tools) - [GrowthBook - open-source feature flagging tools compared](https://www.growthbook.io/blog/best-open-source-feature-flagging-tools-compared) - [Costbench - feature flags pricing benchmarks 2026](https://costbench.com/stats/feature-flags-pricing-2026/) - [Rollgate - feature flags pricing comparison 2026](https://rollgate.io/blog/feature-flags-pricing-comparison)

Markdown edition · the question page, with every recording · Answer Watch · AEO Watch

Answer Watch records what one AI answer engine said on one date, nothing more. Each page names the engine and the model, carries the date the question was asked, and lists the brands that appeared in the answer in the order the engine gave them, with the sources it cited. It is a measurement, not a ranking, not a recommendation and not a claim about the world. Brands that did not appear in an answer are not named on any public page. Coverage: Claude (Anthropic) with web search. Other engines not included.