What GrowthStack is
What is GrowthStack?
GrowthStack is a self-hosted experimentation and feature flag platform: A/B and multivariate tests, feature flags with rollouts and a live kill switch, segments, and event ingestion. It runs on your own infrastructure, and product managers can operate it from Claude in plain English.
Is GrowthStack open source?
No. GrowthStack is built for your company and handed over to you, with the source code included. You own and can modify the code, but it isn't published as a public open-source project.
How is it different from LaunchDarkly, Optimizely or Statsig?
They are cloud services you rent: your flags and experiment data live in their cloud, and pricing grows with users, seats or events. GrowthStack runs inside your own infrastructure, evaluates in the same process as your request with no network call, and is a one-time build rather than a recurring licence.
Why not just self-host a free open-source tool like GrowthBook, Unleash or Flagsmith?
Those are good tools, and their open-source editions are free to start. But they assume your team already has a data warehouse with events flowing in, and engineers to install, integrate and run them, and more advanced features sit in paid per-seat or enterprise plans. GrowthStack is built to your stack and deployed for you, warehouse included, on your own infrastructure. It's handed over with the source code, has no per-seat licence, and includes the Claude layer that answers growth questions on your own data.
Is GrowthStack a complete platform or a lightweight alternative?
It covers the core capabilities the category is expected to have: flag targeting and percentage rollouts, a live kill switch, A/B and multivariate experiments, deterministic assignment, segments, and event ingestion. The
comparison page lays them out side by side.
Deployment, data and security
Where does GrowthStack run?
On your infrastructure: a single VM with Docker Compose, or a full AWS setup with ECS, RDS and ElastiCache if that's what you already run. We deploy it to where you are, not the other way round.
What is it built on?
A Go service backed by your own Postgres and Redis. Rules and segments live in your database, and evaluation happens in the same process as the request.
Does our data leave our infrastructure?
No. Events, experiment assignments, flags and segments are stored in your own Postgres and Redis. Nothing is sent to a GrowthStack cloud, because there isn't one. When your team asks questions in Claude, the answers are computed by tools running on your infrastructure; only the question and the result pass through Claude.
Is it secure and compliant?
Because it runs inside your environment, it sits under your own access controls, network rules and compliance setup. That makes data-residency and GDPR requirements simpler, since no third party holds your experiment data. GrowthStack doesn't hold its own SOC 2 certification; because it runs entirely inside your environment, it falls under your existing security and compliance controls.
How fast is it?
In a published k6 load test, experiment assignment held 10,000 requests a second at 1.3ms p95 and 3.7ms p99, with zero errors across 3.8 million requests. The full method is on the
load test page.
What happens to users if we change an experiment mid-flight?
Assignment is deterministic and sticky: a user who stays in the experiment always sees the same variant. If you reduce traffic, users outside the new allocation leave the experiment, and if they come back later they return to their original variant. Every change is previewed first, showing who stays and who drops out, and nothing happens until you confirm.
Working together
How long does it take?
The core build typically runs 6–10 weeks, in weekly milestones you review: an audit and scoped plan, first experiments and flags live alongside your current tool, event ingestion on your real traffic, then a progressive cutover at your pace.
Can we migrate from LaunchDarkly, Statsig or Optimizely without disruption?
Yes. GrowthStack runs side by side with your existing tool first, so you can check that results match before switching anything off. Cutover then happens feature by feature, not in one big switch. See
how a Statsig migration works.
What do you need from our team?
Access to the environment it will run in, a short walkthrough of your current flags, experiments and events, and someone to review each week's milestone. We agree the time commitment with you upfront, during scoping.
Who supports it after handover?
During the 6–10 week build, we check that every event, experiment, flag and segment works as expected on your real traffic before anything is handed over. After handover, Sameer stays available to help directly if anything fails.
What comes after the core build?
An optional Phase 2 for teams that want more: an advanced statistics engine (CUPED variance reduction, sample-ratio-mismatch detection, guardrail metrics), observability with Datadog dashboards and APM, and AI agents that act on experiment results. It's scoped together once Phase 1 is live, not sold upfront.
Who is behind GrowthStack?
GrowthStack is built by Sameer, who spent 10 years building growth and experimentation systems at large consumer companies (Flipkart, Paytm Mall, Acko, Ola), running experiments across tens of millions of users.
Claude and the AI layer
What can our team do from Claude?
Create, start, edit and pause experiments; create flags, change targeting and rollout, and kill them live; build segments; preview how many customers a rule would reach; check traffic mix; get an explanation for any user's assignment; and define metrics, funnels and retention in plain English, with no engineering ticket.
See the full list.
Can the AI change things without us approving?
No. Every change to a live experiment or flag, including a pause, is shown as a preview first (who is affected and how) and only happens after a person confirms.
Can it answer business questions, not just manage flags?
Yes, from your own event data. You define what an event means (for example, revenue is the amount on checkout_completed) and Claude confirms the definition before saving it, then uses it for funnels, retention and segment questions. Answers state the definition, the time window and what was excluded.
Does the chat write to our data?
No. The chat reads your event data to answer questions; it doesn't ingest or edit events. Test assignments and flag evaluations from chat don't write to the database.
Which AI tools does it work with?
GrowthStack is built and tested for Claude. It connects over MCP (Model Context Protocol), so your team works with it from Claude Desktop.