feature-flags-experimentation
LaunchDarkly open source alternatives
LaunchDarkly starts at $0/mo. Here are 3 open-source alternatives — ranked, opinionated, and refreshed daily against the GitHub API. No paid placements in the rankings. No AI-slop lists.
Comparison table (live data)
GitHub metrics snapshot: 2026-09-08
LaunchDarkly Alternatives
LaunchDarkly Foundation charges $10/month per 1,000 client-side monthly active users. At 10,000 MAU, that's $1,200/year before AI runs, logs, or server-side flags add extra costs. Most teams don't need LaunchDarkly's edge network SLA or enterprise workflow orchestration. They need reliable feature flags that don't scale bills with user growth.
Self-hosting Unleash (AGPL-3.0, 13,728 GitHub stars) or Flagsmith (BSD-3-Clause, 6,508 stars) costs $600/year for unlimited MAU, unlimited seats, and complete control. The server bill is fixed. User growth doesn't touch it.
Quick Comparison
| Usage | LaunchDarkly Foundation | GrowthBook Pro (10 seats) | Unleash (self-hosted) | Savings vs LD |
|---|---|---|---|---|
| 1K MAU | $120/year | $4,800/year | $600/year | $— |
| 5K MAU | $600/year | $4,800/year | $600/year | $— |
| 10K MAU | $1,200/year | $4,800/year | $600/year | $$600 |
| 25K MAU | $3,000/year | $4,800/year | $600/year | $$2,400 |
Prices verified 2026-08-11. LaunchDarkly Developer tier (free) includes only 1K client-side MAU; Foundation is pay-as-you-go beyond that. GrowthBook pricing is seat-based, not MAU-based—you pay for the people running experiments, not the users seeing them. Self-hosted costs assume Vultr Regular 8GB ($480/year) + $120 ops allowance; Unleash can run on smaller instances for lower-traffic deployments.
Why LaunchDarkly Costs $1,200/Year (And Up)
LaunchDarkly restructured pricing in 2024 to eliminate per-seat charges—a welcome change. But the trade-off is usage-based billing that scales with how many users see your flags, not how many team members manage them.
Key pricing components from their 2026-08-11 page:
- Client-side MAU: $10/1,000 per month. This is your web/mobile app users. If your product has 10,000 active users, you pay $1,200/year for Foundation before any other features.
- Server-side requests: Included up to Foundation tier limits, then paid. If you're serving flags to backend services, this meter adds to the MAU bill.
- AI runs: Developer tier includes 5,000 free, then $5/1,000. If you're using Copilot or release automation, this stacks.
- No seat fees is genuine—once you're on Foundation, adding team members doesn't increase the monthly bill. But user growth does.
The pricing model makes sense for LaunchDarkly's target market: enterprises with millions of MAU who value the <150ms config delivery SLA and need approval workflows / SCIM provisioning / audit retention >100 days. For a 10-person team serving 10,000 users, the MAU meter is expensive relative to self-hosting a stateless flag service that evaluates locally.
The Seat-Based Alternative: GrowthBook Cloud
If you're not ready to self-host, GrowthBook Cloud Pro offers a different deal: $40/seat/month ($480/year per person), up to 50 users. Ten seats = $4,800/year, regardless of how many end users see the flags.
GrowthBook's FAQ explicitly contrasts this: "We use a seat-based model rather than charging per monthly active user or API event... you pay for the people who run experiments, not the end users who see them." This works in your favor when user count exceeds team size—common for any B2C product.
Caveats:
- GrowthBook's cloud CDN includes 2M requests/month on Pro, then $10/million. If you're serving flags at high volume, this can add up, though still cheaper than LaunchDarkly's MAU model at scale.
- Enterprise features (SAML SSO, SCIM, approvals, audit logs) are behind a paywall. Self-hosting GrowthBook's open-core gets you everything except those, which may or may not matter depending on compliance needs.
Self-Hosting: Two Strong Options
Unleash (AGPL-3.0)
13,728 GitHub stars (as of 2026-08-11), active development, and the lightest self-host in this comparison.
- Dependencies: PostgreSQL only. No Redis, no ClickHouse, no Kafka. One database, one app server.
- License: AGPL-3.0. Fully open source. If you modify and distribute, you must open-source your changes. If you're just running it internally, there are no restrictions.
- Official Docker Compose: Available, though docs warn "please do not use this configuration for production setups." The compose file is meant for local dev; production deploy recommendations point to managed cloud or multi-instance setups.
- Managed option: Unleash Cloud exists if you want to avoid ops but keep the Unleash UI/SDK ecosystem. Pricing not publicly listed; contact sales.
Minimum server: Official docs don't specify RAM/CPU minimums. Helm chart defaults are empty (resources: {}), with commented examples of 1000m CPU / 768Mi limits. A 4 GB instance (Vultr High Performance $408/year) should handle small-to-medium teams; bump to 8 GB Regular ($600/year) for higher traffic or HA.
✓ Verified: Local deployment test (2026-08-11)
We deployed Unleash OSS locally using the official Docker Compose file. Full startup took 25 seconds (2 containers: unleash-server + postgres:15). Created a test flag via POST /api/admin/projects/default/features, enabled it to the development environment, and verified SDK retrieval via GET /api/client/features with the backend token. All core features (flag creation, gradual rollout strategies, user targeting) worked immediately with zero configuration beyond the default credentials (admin / unleash4all). The claim "just PostgreSQL" is accurate—no hidden dependencies.
Flagsmith (BSD-3-Clause)
6,508 GitHub stars, BSD-3-Clause license (cleanest for commercial use—no copyleft obligations), and the only tool in this comparison with published production sizing guidelines.
- Dependencies: PostgreSQL required. Redis or Memcached optional but recommended for Medium+ deployments. InfluxDB optional for analytics offload.
- License: BSD-3-Clause. No copyleft. Modify, redistribute, commercialize—no strings attached.
- Official sizing: Flagsmith docs break down Small / Medium / Large / XL tiers by component (API workers, DB vCPU/RAM/IOPS, task processor, caching layer). Small starts at 5 vCPU / 14 GB total across components. This is higher than Unleash because Flagsmith documents a production-ready HA setup, not just "minimum to boot."
- Managed option: Flagsmith Cloud pricing not public; contact for quote.
Real cost implication: If you compress Flagsmith Small onto a single instance, you need 8GB+ ($600/year). If you follow their multi-component recommendation (separate API and DB hosts), you're running 2–3 VMs. Still cheaper than LaunchDarkly at 10K+ MAU, but not as light as Unleash.
Cost Breakdown: Where Self-Hosting Wins
| Scenario | LaunchDarkly Foundation | Unleash Self-Hosted | Annual Savings |
|---|---|---|---|
| 5K MAU, 5-person team | $600/year | $600/year | $LD cheaper |
| 10K MAU, 10-person team | $1,200/year | $600/year | $$600 |
| 25K MAU, 10-person team | $3,000/year | $600/year | $$2,400 |
Break-even is around 5,000 MAU. Below that, LaunchDarkly Developer (free for 1K MAU) or Foundation's pay-as-you-go can be cheaper than running a server. Above that, the MAU meter scales linearly while self-hosting stays flat.
Feature Comparison
| Feature | LaunchDarkly | Unleash | Flagsmith |
|---|---|---|---|
| Feature flags (kill switches) | ✓ | ✓ | ✓ |
| Gradual rollouts (% targeting) | ✓ | ✓ | ✓ |
| User/segment targeting | ✓ | ✓ | ✓ |
| A/B testing / experiments | ✓ | ✓ (via metrics integration) | ✓ (via analytics events) |
| Multi-environment (dev/staging/prod) | ✓ (3 on Developer, unlimited on Foundation) | ✓ | ✓ |
| Audit logs | ✓ (14-day retention Developer, 30-day Foundation, 100+ Enterprise) | ✓ | ✓ |
| Approval workflows | ✓ (Enterprise) | ✗ (Cloud only) | ✗ |
| SAML SSO / SCIM | ✓ (Enterprise) | ✗ (Cloud only) | ✗ |
| Edge config delivery SLA | <150ms (Foundation), <200ms (Developer) | No published SLA | No published SLA |
| SDKs | 30+ | 20+ | 10+ |
Bottom line: Core feature-flagging (kill switches, gradual rollouts, targeting) is table stakes—all three deliver it. LaunchDarkly's differentiators are enterprise workflow features (approvals, SCIM, long audit retention) and the edge network SLA. If your team doesn't need those, the feature gap is narrow.
When to Self-Host
- MAU exceeds ~6,000 and growing. This is where LaunchDarkly's meter overtakes a $600/year server. If you're B2C or high-traffic B2B, you hit this fast.
- Team size ≥5 and you have DevOps bandwidth. Unleash/Flagsmith are stateless apps backed by PostgreSQL—standard ops. If you're already running Postgres for your app, the marginal complexity is low.
- You need HIPAA/SOC2 compliance and want data on-prem. Both Unleash and Flagsmith self-host give you full control over where flag data lives. LaunchDarkly's cloud may meet compliance (they have SOC2), but some orgs require on-prem for all config.
- You want seat-free scaling. With self-hosting, adding team members is free. With LaunchDarkly Foundation, it's already free—but user growth isn't. GrowthBook Cloud charges per seat. Pick the axis that scales slower for you.
When to Stay on LaunchDarkly (or Pay for Managed)
- You're under 5K MAU and the free tier covers you. LaunchDarkly Developer is genuinely free for 1K client-side MAU, 10M logs, 5K AI runs. If that's enough, self-hosting costs more (ops time + server).
- You need approval workflows or SCIM provisioning. These are Enterprise-only on LaunchDarkly, but they're not available at all in Unleash/Flagsmith self-hosted. If your compliance framework requires flag-change approvals, you're paying for Enterprise or switching tools entirely.
- You value the <150ms edge SLA. LaunchDarkly's CDN guarantees sub-150ms config delivery on Foundation. Unleash/Flagsmith self-hosted have no such SLA; latency depends on your hosting location and caching setup. For globally distributed apps where flag-fetch latency matters, this is real.
- Ops time is too expensive. If your team's fully loaded cost is high (think $150k+/engineer) and DevOps has a long queue, the $600/year server saving can be negative ROI if it takes 8+ hours/year to maintain. LaunchDarkly offloads that entirely.
Migration Considerations
Switching feature flag platforms is lower risk than switching analytics or email—flags are stateless config, not customer data. But there are steps:
- SDK swap. Unleash and Flagsmith have their own SDKs; you'll rewrite the
if (featureEnabled('new-checkout'))calls. Both have migration guides for LaunchDarkly users. - Flag definitions export. LaunchDarkly doesn't have a one-click "export all flags" button. You'll either script this via their API or manually recreate flags in the new platform. For 10–20 flags, manual is fine. For 100+, script it.
- Targeting rules. User segments and percentage rollouts translate directly. Complex targeting (LaunchDarkly's boolean logic on custom attributes) may need simplification if the new tool's rule engine is less expressive.
- Audit log continuity. Your old LaunchDarkly audit trail stays with LaunchDarkly (retention limits apply). If you need a continuous audit record, export logs before you switch.
Recommended path: Run both platforms in parallel for 2–4 weeks. Deploy the new SDK alongside the old one (reading from different flag services), gradually shift flags over, confirm no regressions, then remove LaunchDarkly SDK and cancel subscription.
Bottom Line
LaunchDarkly Foundation costs $1,200/year for 10,000 MAU before add-ons. That's $<strong>$600 more than</strong> self-hosting Unleash or Flagsmith on a $600/year VPS.
The savings grow linearly with user count. At 25K MAU, you're paying LaunchDarkly $3,000/year; self-hosting is still $600. For a B2C product or high-traffic B2B SaaS, that gap funds a full DevOps headcount within 2–3 years.
This is one of the clearest cost arbitrages in the SaaS-to-self-host category—feature flags are stateless, dependencies are light (just PostgreSQL for both options), and the open-source alternatives are mature and actively maintained. If your team has the DevOps capacity, the $600/year server is a rounding error compared to the MAU meter.
Frequently Asked Questions
Is Unleash really open source, or is it open-core?
Unleash is fully open source under AGPL-3.0. There's no "enterprise edition" directory with paywalled features. Some advanced features (approval workflows, SAML SSO) are available only on Unleash Cloud (their managed offering), but that's a hosting model, not a licensing restriction. You can read the entire codebase on GitHub; nothing is locked behind a commercial license.
What's the catch with Flagsmith's BSD-3 license?
There's no catch. BSD-3-Clause is one of the most permissive licenses: you can modify, redistribute, and commercialize Flagsmith without sharing your changes. The three clauses just require you to preserve the copyright notice and not use the Flagsmith name in endorsements. If you're building a commercial product that embeds feature flags, BSD-3 is cleaner than AGPL because there's no copyleft obligation.
Does self-hosting save money if my team is only 3 people?
Depends on your MAU. LaunchDarkly Developer (free) covers 1K MAU. If you're at 1,500 MAU, Foundation costs $180/year. A Vultr 4GB instance (enough for low-volume Unleash) is $408/year. You save $nothing—LaunchDarkly is cheaper. Factor in ops time: if setup + maintenance takes 8 hours/year and your team's loaded cost is $100k/person ($50/hour), that's $400 of labor. Net savings shrink to $negative.
Rule of thumb: self-hosting starts making financial sense around 5K MAU or when you're already running Postgres and the marginal ops burden is near zero.
Can I use LaunchDarkly SDKs with a self-hosted flag service?
No. LaunchDarkly's SDKs are designed to talk to LaunchDarkly's API. You'll need to swap to Unleash or Flagsmith SDKs (or implement OpenFeature, a vendor-neutral standard that both Unleash and Flagsmith support, plus LaunchDarkly via a provider). The code change is usually small—flag evaluation looks similar across tools—but it's not zero.
What happens to my flags if the self-hosted server goes down?
Depends on SDK caching. Most flag SDKs (including Unleash's and Flagsmith's) cache the last-fetched flag state locally. If the flag service is unreachable, your app continues serving flags from cache until it reconnects. The risk isn't "flags stop working"; it's "flag changes don't propagate." LaunchDarkly's edge network reduces this risk with its SLA, but the failure mode is similar—if their CDN is down, SDKs fall back to cached state. The real question is how long you can tolerate stale flags. For kill switches, minutes matter. For gradual rollouts, hours are fine.
Ready to cut your feature flag bill? Unleash and Flagsmith both offer Docker Compose quickstarts. Spin one up, point a test app at it, and compare the experience. If it feels solid, you're $$600/year ahead.
Ready to deploy?
The self-host options above all run cleanly on modern VPS providers. Our recommended stack:
We earn a commission from DigitalOcean, Cloudways and Vultr when you sign up through the links above, and from some links in the article. We only include hosts we run production workloads on. Prices shown are the vendors' published rates; clicking through does not change what you pay. The OSS rankings above are independent of this section.