Split moved to Harness: stay on Harness FME or replace it
If Split now appears inside Harness on your account, you need to decide before the next contract, and the decision depends on what you are buying. Harness Feature Management and Experimentation (FME) is sold as a module of a quoted Harness Enterprise plan, so its price and terms come from the whole platform agreement rather than from a standalone flag tool. The open-source alternatives are self-hosted GrowthBook or Unleash, which keep flag and experiment data on your own infrastructure. Harness announced its acquisition of Split on May 30, 2024. The choice turns on data residency, licensing, lock-in and cost units, and feature comparisons alone will not settle it.
Start with licensing and exit terms, because they set how long you can depend on the tool. GrowthBook's LICENSE file places code outside its enterprise directories under the MIT Expat license, while those directories fall under the GrowthBook Enterprise License. Unleash moved its server source code to AGPLv3 starting with v8.0.0, while its official open-source Docker images stay under Apache 2.0. Ask your counsel whether your internal modifications or any service you offer to others would trigger source disclosure obligations. On the cost side, Harness prices FME by quotation, while Unleash lists Pay-As-You-Go at $75 per seat per month with a 5-seat minimum for self-hosted deployments. Ask Harness to itemize FME separately from the rest of the Enterprise plan so you can compare it line by line.
Before renewal, get written answers from Harness on four points: which modules and seat counts the quote includes, the date your tenant moves to app.harness.io, where your flag and event data is stored, and which export formats are available. Then check your own dependencies. If your design uses Unleash Edge as a caching layer, note that the open-source version reaches end of life on December 31, 2026. Use the answers to score each option against your constraints, and keep a parallel-run plan ready so you can switch providers without a single cutover date forcing the decision.
Which requirements should drive a stay or move decision
Split's move into Harness changes the contract, the hosting model and the exit path at the same time, so a renewal decision needs more than a feature comparison. The cards below set out the requirements that should drive the choice, and each one points to a fact you can verify before you sign anything.
Where flag and experiment events live
Ask where exposure events, metric events and experiment results are stored and who can query them. GrowthBook analyzes experiments against your own warehouse, including BigQuery, Snowflake, Redshift and Postgres, and runs only SELECT queries that return aggregated statistics. Harness FME is a hosted service on app.harness.io after migration day, so confirm with Harness which event data leaves your environment, how long it is retained and where it is processed.
Licensing terms of the open-source option
Open-source does not mean one license across the product. GrowthBook is MIT Expat licensed outside its enterprise directories, which are under the GrowthBook Enterprise License, so your legal team should review the directories that hold the features you need. Unleash moved its server source to AGPLv3 with v8.0.0, while the official Docker images stay under Apache 2.0. Confirm how that change applies if you ever modify the server or offer it to others.
Pricing unit and quote scope
Harness FME is sold as a module of a Harness Enterprise quote, so the cost depends on the whole Harness contract rather than on flags alone. GrowthBook and Unleash publish per-seat paid tiers, and the self-hosted tiers still carry infrastructure, database and staff costs. Ask Harness for the module price, the platform modules bundled into the quote, the renewal uplift and any usage thresholds, and compare that with per-seat tiers for the same number of users.
Project and environment limits
Free editions set hard boundaries that matter for a multi-team Split estate. Unleash Open Source allows 1 project and 2 environments, while GrowthBook's self-hosted Open Source plan allows 1 project with unlimited users. Count the projects, environments and flags you run today, including staging and regional copies, and check whether they fit the free tier. If they do not, the decision becomes a paid tier decision, not a free migration.
SSO, SCIM and approval workflows
Identity and change control often sit outside the flag tool itself, so list which controls auditors and security reviewers expect. Self-hosted GrowthBook Enterprise adds SSO and SCIM, advanced permissioning, approval workflows and safe rollouts that the free self-hosted edition does not include. Unleash Enterprise adds single sign-on, SCIM provisioning, project roles and change requests. Map each control to the edition that provides it before you compare totals.
Exit effort and parallel running
Whatever you choose, plan the exit from Split before the contract ends. GrowthBook has a LaunchDarkly importer and a Statsig importer, but its documentation warns that advanced setups may include cases it cannot handle, and the application still needs an SDK switch after import. Unleash recommends exporting definitions as JSON, recreating them through its Admin API and running both systems in parallel behind an abstraction layer before cutover. Ask for a written export of your Split definitions before you give notice.
Which path fits your team: stay, GrowthBook or Unleash
Teams that bought Split mainly to run A/B tests have the closest match in GrowthBook, which handles feature flags and experiment analysis in one tool. Experiments are read from your own warehouse, so the analysis logic moves with your data rather than staying inside a vendor's hosted metrics.
- Warehouse analysis: GrowthBook analyzes experiments against sources such as BigQuery, Snowflake and Postgres, running only SELECT queries and transmitting only aggregated statistics.
- Free self-hosted limits: The self-hosted Open Source plan allows 1 project with unlimited users, so count your Split projects before you commit to it.
- Paid self-hosted features: SSO, SCIM, approval workflows, safe rollouts and holdouts sit in the self-hosted Enterprise tier.
- Metric rebuild: Hosted Split metrics must be rebuilt as queries on your warehouse, which needs data engineering time before the first result.
Confirm which metrics your running experiments depend on and whether your warehouse can serve them before choosing this path.
Teams that use Split mostly for release flags and gradual rollouts will usually find Unleash the closer operational fit. Unleash is a flag management tool, so experiment analysis would need a separate product alongside it.
- Open Source limits: Unleash Open Source allows 1 project and 2 environments with basic roles, so a multi-team estate will need Unleash Enterprise.
- Licensing change: From v8.0.0 the server source code is AGPLv3, while the official open-source Docker images stay under Apache 2.0. Unleash states that anyone who modifies it to offer as a public or commercial service must publish the source or buy a commercial license.
- Edge planning: Open-source Unleash Edge reaches end of life on December 31, 2026, so scaled designs that rely on it need a replacement caching plan.
- Separate experiment tool: Plan a second product for experiments if they are part of your current Split use.
Count your current Split projects and environments, then confirm which Unleash edition covers them and how your Edge design will change.
Regulated teams that must keep flag payloads, targeting data and experiment results inside their own environment should compare the self-hosted options directly with the hosted Harness FME service. Both GrowthBook and Unleash run on infrastructure you control, but the components you operate differ.
- GrowthBook deployment: The front end, API and statistics engine ship in one Docker image, with an official Helm chart for Kubernetes and metadata kept in MongoDB or a compatible database.
- File storage choices: GrowthBook file uploads can sit on local storage, Amazon S3 or Google Cloud Storage, so you decide where those objects live.
- Unleash deployment: The unleashorg/unleash-server image runs against PostgreSQL 15 or later, with the server and admin UI on port 4242.
- Residency confirmation: Ask Harness in writing where FME stores flag and event data, and compare that answer with your policy.
Map each data flow, including SDK calls, metadata, event data and uploads, to a location your policy allows before you compare options.
Staying on Harness FME makes sense when your organization already buys Harness Enterprise, when Split SDK integrations are deep in your applications, and when one vendor relationship is a priority. The trade-off is that cost follows the quoted Harness Enterprise plan, so the figures come from the contract rather than a published list.
- Module pricing: FME is sold as a module of a quoted Harness Enterprise plan, so ask for a written quote that names the modules, seats and features included.
- Migration day: Accounts move to app.harness.io on migration day, so review SSO configuration, bookmarked URLs and API integrations ahead of that date.
- Platform consolidation: If your organization already uses other Harness modules, one console and one contract may reduce administration.
- Exit options: Staying does not remove the option to leave later, so confirm with Harness in writing which export options exist for flag definitions.
Request the quote and export terms before the renewal date so you can compare them directly with the self-hosted paths.
How Harness FME, GrowthBook and Unleash compare
This table compares Harness FME, self-hosted GrowthBook and self-hosted Unleash on data control, licensing, pricing, free-edition limits, self-hosting components and migration tooling. Read the Harness column as a list of items to confirm in your quote, since its scope depends on your contract. The main takeaway is that the two open-source options differ in edition limits, so the free tier you choose affects what you can keep.
| Criterion | Harness FME | GrowthBook | Unleash |
|---|---|---|---|
| Data control | Hosted by Harness at app.harness.io; confirm where flag and event data is stored and whether any component can run in your environment | Experiments query your own warehouse with SELECT only and receive aggregated statistics | Self-hosted server and database in your infrastructure; flag data stays on your PostgreSQL instance |
| Licensing | Commercial subscription under the Harness agreement; confirm terms in the quote | MIT outside the enterprise directories; those directories fall under the GrowthBook Enterprise License | AGPLv3 for source from v8.0.0; official Docker images remain under Apache 2.0 |
| Pricing unit | Quoted as a module of the Harness Enterprise plan; ask which units the quote uses | Free self-hosted tier; paid tiers per seat, with Cloud Pro at $40 per seat per month (GrowthBook pricing page) | Self-hosted Pay-As-You-Go at $75 per seat per month with a 5-seat minimum (Unleash pricing page) |
| Free-edition limits | No free self-hosted edition is documented here; confirm any trial or evaluation terms | Open Source plan has unlimited users, 1 project and unlimited API requests, but no SSO, approvals or holdouts | Open Source has 1 project, 2 environments and 5,000 feature flags per instance, with basic roles only |
| Self-hosting components | Vendor-operated service; ask Harness what, if anything, runs in your infrastructure | One Docker image with front end, API and statistics engine, plus MongoDB or a compatible database and an official Helm chart | unleash-server image on PostgreSQL 15 or later, port 4242; Unleash Edge is needed for caching at scale |
| Migration tooling | Confirm which export and import methods Harness provides for flags, segments and experiments | Built-in LaunchDarkly and Statsig importers; the app still needs an SDK swap after import | JSON import and export, a LaunchDarkly migration guide and Admin API recreation of segments and custom strategies |
| Edge and caching | Not applicable to a hosted service; confirm how Harness handles SDK traffic at your scale | Not applicable to GrowthBook as documented here; confirm caching needs with its SDK documentation | Open-source Unleash Edge reaches end of life on December 31, 2026; plan an Enterprise plan or another cache |
How to decide and deploy before the renewal date
This sequence turns the renewal decision into a tested cutover: you inventory what Split actually runs, export its definitions, stand up a self-hosted server, and only switch production traffic once both systems return matching results. Each step ends with a check you can show to finance and security before the contract date.
Inventory flags, segments and experiments
Start with a list of every feature flag, segment, experiment and metric that production and analytics depend on, including which applications call each one and which environments they run in. Platform or product owners should own this, with input from the teams that maintain the SDK integrations. Note which items are release flags, which are kill switches, and which are experiments with results that someone still reads. The step is done when every item has an owner, a consuming service and a retire-or-keep decision, so you know the real migration scope before you compare tools.
Export definitions from Split or Harness FME
Pull flag definitions, treatments, targeting rules and segments out of Split or Harness FME through the Admin API, using the export tooling Harness documents for its platform. Store the JSON or YAML outputs in version control with a dated snapshot, so the export is reproducible if the tenant changes during the renewal window. Expect some targeting rules and segments to need manual recreation, which you should record as a separate list. The step is done when a second engineer can open the export, count the flags against your inventory, and find no unexplained gaps.
Decide between GrowthBook and Unleash
Choose GrowthBook if experiments and warehouse-native analysis matter, because it runs read-only SELECT queries against your own data source and sends only aggregated statistics back. Choose Unleash if you mainly need release flags and rollouts. Check the license of the edition you plan to run: Unleash v8.0.0 moved the server source to AGPLv3 while the official open-source Docker images remain under Apache 2.0, and GrowthBook code outside its enterprise directories is MIT. Ask legal to confirm both readings for your use. The step is done when a written decision records the tool, the edition, the license reading and which governance features you need from a paid tier.
Stand up the self-hosted server
For GrowthBook, deploy the single Docker image or the official Helm chart, backed by MongoDB or a compatible database, and consider a managed service for production. For Unleash, run the unleashorg/unleash-server image against PostgreSQL 15 or later, with the server and admin UI on port 4242. Your platform team sets up backups, TLS, network placement and SSO where the chosen edition allows it. Use the vendor's getting started and self-hosting documentation as the reference. The step is done when a health check passes in a non-production environment, backups restore into a test copy, and the admin login follows your identity policy.
Import or recreate definitions
Load the exported definitions into the new system. GrowthBook has a documented Split migration path that reads environments, segments, feature gates, dynamic configs and experiments with a Console API key, though its metric support is limited and in beta. Unleash supports environment import and export as JSON, but segments and custom strategies must already exist in the target before import. Recreate any remaining rules by hand from your manual list. The step is done when each flag in the inventory exists in the target with its environments, targeting and default values, and the gap list is closed or accepted by its owner.
Run both systems in parallel behind an abstraction
Put an abstraction layer between application code and the flag vendor, such as the OpenFeature API with a provider for each system, so services can switch providers without code rewrites. Evaluate the same flags in both systems for a sample of traffic and compare the returned values per user or request. For experiments, rebuild the metrics on your warehouse and confirm that the new results match the old ones within the tolerance your analysts agree on. The step is done when mismatches are explained, each mismatch has a fix or an accepted reason, and the analysts sign off on the metric definitions.
Cut over and retire Split
Switch production flag evaluation to the new provider one service or environment at a time, starting with low-risk flags, and keep the Split configuration readable until the last service has moved. Confirm the SDK swap in each application, record the cutover date, and check that alerts and dashboards reflect the new system. Only after a stable period, and after the data retention obligations you agreed with legal are met, should you cancel the Harness FME or Split subscription and remove credentials. The step is done when no production service calls the old SDK, the final export is archived, and the renewal record documents the decision.
What do buyers ask before moving off Harness FME?
What does it cost to move off Harness FME?
The cost depends on how many flags, segments and experiments you run, and on who does the work. Harness FME is quoted as a module of a Harness Enterprise plan, so your exit cost is mostly internal labor, the self-hosted infrastructure you choose, and any paid open-source tier you need for SSO, approvals or governance. Ask Harness for the current FME and Enterprise terms in writing, and ask your Harness account team how the module is billed at renewal. For the open-source side, the GrowthBook pricing page lists a free self-hosted Open Source plan with 1 project and unlimited users, and Unleash lists a free self-hosted Open Source plan with 1 project, 2 environments and 5,000 feature flags per instance. Recurring costs are the hosting, the database, and the paid seats if you need SSO or approval workflows.
Which cost components drive a move from Split to a self-hosted tool?
Four components drive most of the cost. The first is one-time migration work: exporting definitions, recreating targeting and segments, and switching SDKs in each application. The second is infrastructure you run yourself, including compute, a production database and backups. The third is warehouse query cost for experiments, since GrowthBook runs read-only SELECT queries against your own warehouse. The fourth is paid seats for enterprise features such as SSO, SCIM and approval workflows, which the free self-hosted editions do not include. Each component is priced by a different party, so ask your quote to separate them.
How long does a migration off Harness FME take?
The timeline depends on the number of flags and experiments, how many applications use the SDK, how complex the targeting is, and whether you need to run both systems in parallel. No supplied source gives a duration for a Split or Harness FME migration, so ask for a timeline in the written scope rather than relying on a generic estimate. The GrowthBook LaunchDarkly importer documentation warns that advanced setups may include scenarios it cannot handle, which is a useful sign of where extra time goes. Ask for the scope to list the flags and experiments in scope and the count of applications to re-instrument.
What must our internal team provide during a migration?
Your team has to supply access, decisions and verification. Expect to provide admin access to Harness and to the target platform, a list of which flags, segments and experiments are business-critical, owners who can approve targeting logic, and engineers who can change each application's SDK calls. You also need a data or analytics owner to confirm that warehouse metrics match the results you trust today. Ask the implementation partner to name these roles in the written scope so that approvals and test windows are agreed before work starts.
Can our own engineers run the migration without a partner?
Yes, a team with platform and SDK experience can run the migration in-house. Unleash documents an export of flag and segment definitions as JSON, and recreating them through its Admin API or import feature, and GrowthBook documents its own importers. The risk is in the details: segments and custom strategies must exist in the target before import, and the application must switch SDKs after the import. Running both systems in parallel behind an abstraction layer, such as OpenFeature, reduces cutover risk. Check whether your team has capacity for parallel running and for an audit trail of changes during the cutover.
Does DeployOpen handle the whole migration and how is the work scoped?
DeployOpen plans, implements, customizes, integrates, hardens, backs up, upgrades, documents and supports self-hosted open-source platforms inside your own infrastructure. Most engagements are fixed-scope, with scope, deliverables and timelines agreed in writing before work starts, and ongoing support is available as a retainer. DeployOpen does not publish a price list, so a quote follows from the written scope. DeployOpen does not issue audit reports or certifications, so any compliance evidence still goes to your independent auditor.
How do GrowthBook and Unleash licenses affect our plans?
The two licenses fit different uses. GrowthBook's LICENSE file says code outside its enterprise directories is MIT Expat licensed, while those enterprise directories fall under the GrowthBook Enterprise License, so paid features are not open source. Unleash moved its server source from Apache 2.0 to AGPLv3 with v8.0.0, while its official open-source Docker images stay under Apache 2.0. Unleash's blog states that running the official images without modification does not trigger the change, but modifying Unleash and offering it as a public or commercial service requires publishing the source or buying a commercial license. Have your legal team review both licenses against your planned use before you commit.
What should we ask Harness for before the renewal decision?
Ask Harness for the FME module pricing as a line item of your Harness Enterprise quote, the migration path or export options available to you, and the notice period and renewal terms for your existing Split contract. Also ask how and when your accounts move to app.harness.io, what changes in the SDKs and data retention, and which features sit in FME versus other Harness modules. Compare the written answers against your flag count, experiment volume and required governance features before deciding.
What auditors and procurement ask for under this decision
Whichever way the decision goes, the same evidence set shows what you buy, where flag and experiment data sits, and who can change production flags. Assemble it before the renewal conversation, not during the audit.
Written Harness quote and module scope
Ask Harness for a written quote that names FME as a module of your Harness Enterprise plan, with the modules, seat or usage units, environments and support terms listed separately. Procurement will compare this against any self-hosted option, so confirm whether FME features you use today (experiments, metrics, approvals) are included or priced as add-ons. Record the quote date and validity period. Harness pricing is quote-based, so this document is the only price basis you will have.
License files and edition comparison
For each candidate, record the exact license and the edition you would run. GrowthBook's MIT Expat terms apply outside its enterprise directories, which are under the GrowthBook Enterprise License, so a self-hosted Enterprise build is a different legal position from the Open Source plan. Unleash moved its server source to AGPLv3 in v8.0.0 while its official Docker images stay under Apache 2.0. Ask counsel to review the terms that match the build you actually deploy.
Data flow diagram for flag and experiment events
Draw where exposure events, flag evaluations and experiment metrics are created, transmitted and stored. In a hosted Split or Harness setup, evaluation traffic and event data leave your network. In a self-hosted GrowthBook design, the application, the GrowthBook service, its MongoDB-compatible database and the warehouse queries each become a box with a stated location. Auditors will ask which personal data appears in events and how long it persists.
Access control and SSO configuration
Document who can create, change and approve production flags, and how that is enforced. Check which SSO, provisioning (SCIM) and approval workflow features each candidate includes in its paid or enterprise tier, since the free self-hosted editions omit several of them. Export the role matrix and the identity provider group mapping as evidence. Test that a leaver loses flag-editing rights on the same day their directory account is disabled.
Warehouse query permissions
If experiments run against your warehouse, the evidence is the database role and the query log. GrowthBook documents that it runs only SELECT queries and transmits aggregated statistics, which is a claim to verify, not assume. Keep the read-only role definition, the list of tables it can read, and a sample of executed queries. This is often the first control a reviewer asks to see for experimentation tooling.
Migration parity test results
Before cutting over, run both systems side by side and record that each flag returns the same value for the same test user across targeting rules, rollouts and prerequisites. Keep the test cases, the results and the signed-off exceptions, such as rules the importer could not carry over. Unleash recommends running both systems behind an abstraction layer before cutover. These records show a change-controlled move rather than an undocumented swap.
How DeployOpen runs a Split exit or stay review
DeployOpen starts with a written review of your Split or Harness FME use, then decides with you whether staying on Harness FME or moving to a self-hosted platform makes sense at your next renewal. If you move, we implement, test and document the replacement inside your own infrastructure. Both paths begin with the same scope.
How we scope
- We inventory which Split features you use: flags, experiments, metrics, segments and SDK types.
- We record your renewal date, Harness contract terms and any SSO, approval or audit needs.
- Engagements are fixed-scope, with deliverables and timelines agreed in writing before work starts.
How we build
- We map each Split targeting rule and experiment to the candidate platform and flag gaps.
- We stand up the chosen platform in your cloud account, data center or air-gapped network.
- We test flag evaluation, rollouts and experiment analysis against your existing results before cutover.
How we hand over
- We deliver written runbooks, configuration records and the evidence needed for your internal reviews.
- We document the SDK switch, the parallel-run period and the decommissioning steps for Split.
- We walk your engineering and compliance teams through operations, backups and upgrade procedures.
How we support
- Ongoing support, upgrades and continuous testing are available as a retainer after handover.
- Retainers can cover platform upgrades, backup checks and periodic security testing of the flag service.
- We confirm current license terms and edition limits with you before each upgrade decision.
