LaunchDarkly

1. 1

   You create a project with environments and define flags. LaunchDarkly's evaluation model is built around contexts: a context is any entity you want to target, a user, an organisation, a device, a request, each with attributes, and flags are evaluated against one or several context kinds at once. That multi-context model is more expressive than the single user identity most competitors use, and it is how B2B products target both the account and the person inside it.

2. 2

   Applications install an SDK. Server-side SDKs for Java, Python, Node, Go, Ruby, Rust, PHP, .NET and others stream the flag configuration and evaluate in process; client-side and mobile SDKs cover JavaScript, React, iOS, Android, and Flutter; edge SDKs run on Cloudflare, Vercel, Akamai, and Fastly; and OpenFeature providers exist for teams coding against the vendor-neutral standard. A self-hosted Relay Proxy sits in front of the fleet when you want one connection out of your network rather than thousands.

3. 3

   Targeting rules decide which context sees which variation, with percentage rollouts, prerequisite flags, and segments. Progressive rollouts ramp traffic on a schedule, and on Enterprise the Guardian add-on watches release health metrics and can roll a release back automatically when a guardrail moves the wrong way.

4. 4

   Experiments attach metrics to flags. Metrics can come from LaunchDarkly's own event stream, from click and page view tracking, or warehouse-native from Snowflake, BigQuery, Databricks, or ClickHouse, so the numbers can be computed against the same tables your data team already trusts. Results are presented as frequentist confidence intervals or Bayesian credible intervals, and bandit experiments shift traffic toward winners rather than holding a fixed split.

5. 5

   Since the Highlight.io acquisition, the same SDK surface also produces session replays, error reports with mobile stack trace symbolication, logs, traces, and LLM observability spans, with metric exemplars linking a point on a chart back to the exact trace and session that produced it. That link between a metric spike and a replay is the strongest argument for buying the bundle rather than three tools.

Freemium and pay-as-you-go, metered on client-side monthly active users and service connections with unlimited seats on all plans, plus usage allowances for session replays, errors, logs, traces, and AI runs. Enterprise is quoted.

Value assessment: The free Developer tier is the best value in this entire category and it is not close: unlimited flags, unlimited seats, 100,000 experimentation monthly active users, plus replay, errors, logs, and traces, for nothing. Beyond it the arithmetic is legible. At 10,000 client-side monthly active users, Foundation is roughly $83 a month at $8.33 per thousand, plus service connections beyond the five included. At 100,000 client-side monthly active users it is roughly $833 a month on the same basis, before observability overage. Compare that with Unleash at $75 per seat, where a ten-person team pays $750 regardless of scale, and LaunchDarkly is cheaper for large teams on small products and more expensive for small teams on large ones. The real value question is not the meter, it is the Enterprise cliff: approvals, custom roles, SCIM, and long retention are not purchasable with a card, so a company that needs governance loses the self-serve pricing advantage entirely and enters a negotiation with a $3B vendor.

Setup time

Under an hour for a first flag. Create a project, install an SDK, wrap a code path, ship. The observability half takes longer because instrumenting logs, traces, and replay properly means touching more of the application, and the Relay Proxy is a separate deployment decision.

Learning curve

Low to start and steep at the edges. Basic flagging is immediately intuitive. Contexts and context kinds take a while to model well, and modelling them badly early produces targeting rules that are painful to unpick later. The experimentation and observability surfaces each carry their own concepts on top.

Onboarding

Fully self-serve on Developer and Foundation, with a 14 day Foundation trial. Enterprise requires sales engagement, and enterprise trials are available on request. Documentation is extensive and covers every SDK, the Relay Proxy, warehouse-native experimentation, and Data Export.

Migration notes

OpenFeature provider support means code written against the neutral standard can point at LaunchDarkly without rewriting call sites, and the same applies in reverse if you leave. Flags and segments are recreatable through the REST API or the Terraform provider. The genuine lock-in risk is on the observability and experiment history side, since replays, traces, and results accumulate in the platform and retention beyond 100 days is an Enterprise term. Data Export to Snowflake, BigQuery, Databricks, or ClickHouse is the correct hedge and should be configured from the start.

Platforms

JavaPythonNodeGoRubyRustPHP.NETJavaScript and ReactiOSAndroidFlutterCloudflare, Vercel, Akamai, and Fastly edge runtimesOpenFeature providersRelay Proxy

API

Full REST API, Terraform provider, Data Export to Snowflake, BigQuery, Databricks, and ClickHouse, a Snowflake Native App for warehouse-native experimentation and analytics, plus webhooks and an extensive integration catalogue.

Compliance

SOC 2GDPRConfirm the current certification list including ISO 27001 and HIPAA with the vendor during procurement

Data residency

LaunchDarkly is a hosted service with no self-hosted option at the self-serve tiers. The Relay Proxy keeps SDK connections inside your network and reduces outbound traffic, but the control plane and stored data remain with the vendor. Buyers with hard residency requirements should discuss options with sales or consider Unleash or Flagsmith instead.

SSO

SSO is available on paid plans; SCIM provisioning for automated user lifecycle management is an Enterprise feature.

Security notes

Server-side SDKs stream configuration and evaluate in process, so targeting attributes can stay in your application rather than being sent per request. Approval workflows, custom roles, and team-scoped permissions provide change control but require an Enterprise contract. Data retention beyond 100 days is also an Enterprise term.

Channels

Documentation and community support on DeveloperStandard support on FoundationDedicated support and success management on Enterprise

Documentation

Extensive documentation at launchdarkly.com/docs covering every SDK, contexts and targeting, experimentation, the Relay Proxy, warehouse-native experimentation, Data Export, and the observability, replay, and error monitoring products acquired from Highlight.io.

Community

Large developer community built up over a decade, a well-known technical blog on release engineering practice, and an active presence in the OpenFeature standardisation effort.

Founded

2014

Headquarters

Oakland, California, United States

Ownership

Venture-backed, privately held

Founders

Edith Harbaugh, John Kodumal

Employees

Approximately 620 to 650 (2026)

Funding

Approximately $329M raised, most recently a $200M round in August 2021 at a $3B valuation.

LaunchDarkly spent a decade as the expensive enterprise answer and then shipped the most generous free tier in its category, which makes it worth evaluating even if you assumed it was out of reach. Unlimited seats on every plan, unlimited flags, 100,000 experimentation monthly active users, and real observability allowances at zero cost is a serious offer, and the pay-as-you-go Foundation tier is easy to forecast at roughly $83 a month per ten thousand client-side monthly active users. The multi-context targeting model and the SDK breadth are the best in the category, and the Highlight.io integration produces genuinely useful workflows that flag-only vendors cannot match. The reservations are structural rather than cosmetic. There is no self-hosting, the statistical methodology is less transparent than its open source rivals, and the moment you need approvals, custom roles, SCIM, or long retention you leave self-serve pricing and enter a negotiation. Start on the free tier, model your client-side monthly active users honestly, and know in advance where the Enterprise cliff sits relative to your compliance roadmap.

Written by the SaaSTracker editorial team. Awards, when shown, are judged against the published criteria in our methodology.