LaunchDarkly
The runtime control layer for feature releases, experimentation, observability and AI agents.
By LaunchDarkly · HQ Oakland, US · 4.5/5 Value-Position score (estimate)
Positioning guardrails
Best for
- Engineering teams releasing frequently who want progressive rollouts, targeting and instant rollback
- Platform teams that need one governance layer for flags, approvals, audit logs and RBAC across many services
- AI product teams that must configure, evaluate, release and observe agent behaviour in production
- Growth and product teams running A/B/n tests and warehouse-native experiments against Snowflake, BigQuery or Databricks
- Enterprises with FedRAMP, SOC 2 or other regulated release requirements
Ideal size: 20–1,000+ people · Scale-up or enterprise with continuous delivery and a platform or DevOps function
Not for
- Teams that release rarely and have no need for runtime control
- Buyers looking for a cheap single-seat flag tool with no governance
- Non-technical organisations with no engineering capacity to integrate SDKs
- Companies that want no-code A/B testing without any developer involvement
Value metrics scorecard
Time-to-Value
1–2 weeks
~14 days to first production value
Total Cost of Ownership
$15,000/yr
Starts at $0 · Usage-based. Free Developer tier; Foundation pay-as-you-go at $10 per service connection/mo plus $8.33 per 1k client-side MAU/mo; Enterprise custom. No platform or per-seat fees.
Implementation Friction
2/5
Engineering + admin effort required
Value-Position score
out of 5 · model estimate
Full cost breakdown
Mandatory implementation fee
None
Seat tiers
Unlimited seats on every plan; billing is based on usage rather than seats.
Add-on costs
- Guardian release-safety add-on on top of Enterprise (monitoring, guardrail metrics, automatic rollback), priced separately
- $5 per additional 1,000 AI runs beyond the 5,000/mo included on Foundation
Company & support
Who is behind LaunchDarkly, and how your team gets help once it is live.
Company
- Founded
- 2014 · 12 yrs in business
- Headquarters
- Oakland, US
How you get support
- PhoneNot listed
- EmailNot listed
- Live chatNot listed
- Support portal / ticketsPlan not stated
- Community forumAll plans
- Help centre / docsAll plans
- Dedicated account managerEnterprise only
- In person / on-siteNot listed
- Hours
- Business hours
- Response time
- Typically within 2 hours (Mon–Fri, 9am–6pm PST); 1 business day response commitment on Enterprise
Support levels are Bronze on Developer and Foundation and Silver on Enterprise, with Gold and Platinum upgrades available. Customer Success Managers, advisory sessions and a guided onboarding programme are Enterprise or add-on services.
“Not listed” means the vendor’s public pages don’t mention that channel, not that it is unavailable. Ask about it during evaluation.
Market position
Where LaunchDarkly sits against its closest alternatives. Pick any two of cost, speed, friction and buyer score, and up to 9 companies to compare.
Quadrant view
Typical annual cost × Time-to-value
The lines cross at the median of the solutions shown, so about half sit on each side of each line.
Companies on the chart 6 / 10
- LaunchDarkly
- Datadog
- Hex
- Appsmith
- Estuary
- Metabase
Add or change companies
Up to 10 companies including LaunchDarkly. Listed closest first.
Stack fit signal
Compatibility with standard B2B ecosystems.
Ships an official MCP server. Connects to Claude Code, Claude Desktop, ChatGPT connectors and Cursor out of the box.
AI & MCP readiness
What LaunchDarkly ships in AI, and what it asks of your ecosystem.
AI features shipped
AgentControl centralises prompts, models and tools, adds an LLM playground, offline evals on golden datasets and online evals with built-in or custom judges, then releases agent changes with guarded rollouts, adaptive triggers and guardrails. Agent graphs, traces and AI Insights cover observability; RBAC and audit trails govern agent configuration.
Your data & models
- Trains on your data
- Not recorded — ask the vendor
- Runs on
- OpenAI, Anthropic
- AI pricing
- Billed by usage or credits
In your ecosystem
- AI connection
- Official MCP server
- Model key
- Not recorded
- AI usage audit
- Not recorded
Compliance attestations
* IAPP AIGP certifies individuals, not products. It means named staff hold the credential — not that the platform does.
Bottom line
LaunchDarkly is a runtime control platform for software releases and AI agents. It started in 2014 as a feature-flag service and now bundles progressive rollouts, automated rollback, observability, warehouse-native experimentation, and AgentControl for configuring, evaluating and governing AI agents in production. Pricing is usage-based with a free Developer tier, a pay-as-you-go Foundation tier and custom Enterprise plans; a Guardian release-safety add-on is sold separately. It reports 5,500+ customers including roughly a quarter of the Fortune 500.
Frequently asked questions
How long does implementation take?
LaunchDarkly is self-serve: the Developer tier is free and you can connect an SDK and evaluate your first flag in hours. Production rollout across many services, plus governance (RBAC, approvals, audit logging) and experimentation setup, is normally a staged project. The vendor publishes quickstarts for server-side, client-side, edge and AI SDKs, plus an onboarding skill for coding assistants.
What does LaunchDarkly cost?
The Developer plan is free forever with unlimited seats and 5,000 AI runs per month. Foundation is pay-as-you-go: $10 per service connection per month and $8.33 per 1,000 client-side MAU per month, with 5,000 AI runs included and $5 per additional 1,000. Enterprise is custom priced with advanced targeting, workflows, SCIM and longer retention; Guardian release safety is a separate paid add-on.
Does LaunchDarkly use customer data to train AI models?
The vendor pages we reviewed do not state whether customer data, prompts or traces are used to train AI models, so no definitive answer can be given. LaunchDarkly's security page says PII is not required to be passed to the service and that it runs its privacy programme in line with GDPR and CCPA. Ask the vendor for a contractual answer covering prompts, traces and eval data before adopting AgentControl.
Which security certifications does LaunchDarkly hold?
Its security page states SOC 2 Type II, ISO 27001, FedRAMP authorization and Data Privacy Framework participation, and says its protections let customers use the products in compliance with HIPAA and FedRAMP. The HIPAA wording concerns enabling customer compliance rather than a vendor HIPAA certification. No ISO 42001 AI management certification is claimed on the pages we reviewed.
Can LaunchDarkly govern AI agents as well as feature flags?
Yes. AgentControl sits alongside the flag product and covers prompt and model configuration, an LLM playground, offline and online evals, custom judges, guardrails with PII redaction, agent graphs, LLM traces, A/B testing of configurations and automatic failover between model providers. Usage is metered as AI runs on every plan.
What support is included?
Developer and Foundation include Discord community support plus the knowledge base and help centre. Enterprise adds a one-business-day response commitment, Customer Success Managers, advisory sessions and a guided onboarding programme, with Gold and Platinum upgrades available. The contact page quotes a typical two-hour response during Monday to Friday business hours, US Pacific time.