Is self-hosted Langfuse cheaper than Braintrust Pro?
Self-hosting Langfuse tends to cost less than Braintrust Pro only when your team already runs ClickHouse, PostgreSQL and object storage and can absorb the operating work. Braintrust Pro is published at $249 a month, but processed-data, score and retention overages and any Enterprise contract are set per agreement, so the answer depends on your invoices. Ask Braintrust for the overage rates that apply to your projected trace volume, score count and retention window, then size a Langfuse deployment for the same volume so both sides of the comparison use identical numbers.
Langfuse's core features are MIT licensed, and self-hosted deployments use the same codebase as Langfuse Cloud with no feature limits, so the software itself carries no license fee. The recurring cost moves into infrastructure. A self-hosted Langfuse deployment runs Langfuse Web and Langfuse Worker alongside PostgreSQL, ClickHouse, Redis or Valkey and S3-compatible storage. Langfuse's scaling guide lists minimums of 2 CPU and 8 GiB for ClickHouse, with at least 16 GiB of ClickHouse memory for larger deployments, and 1 CPU and 1.5 GiB for Redis or Valkey. Your cloud bill or hardware budget covers these services, along with backups and the extra ClickHouse disk needed during upgrades and backfills.
Enterprise license keys add cost for some features, including project-level RBAC roles, data retention policies, audit logs, server-side data masking and SCIM provisioning. Langfuse's self-hosted pricing page describes a custom-priced Self-Hosted Enterprise plan with a named support engineer and a support SLA, so request that quote in writing if you need those controls. The largest unknown is staff time: who patches ClickHouse, runs the v3 to v4 upgrade that requires ClickHouse 25.12 or later, and carries the on-call load. Compare the Braintrust contract with the self-hosted total, including loaded engineering time, before deciding.
Which cost components change when you self-host Langfuse
This table lists each cost component on the Braintrust side and the self-hosted Langfuse side, who is paid for it, whether it is one-time or recurring, and what drives it up. Read the Braintrust rows as what you pay today and the Langfuse rows as what your team takes on, then use the drivers to ask for a written quote. Most of the Langfuse cost is recurring infrastructure and people, not licensing.
| Cost component | Who is paid | One-time or recurring | Main cost driver |
|---|---|---|---|
| Braintrust Pro subscription | Braintrust | Recurring | Published Pro plan price and the plan tier you select |
| Braintrust usage overages | Braintrust | Recurring | Processed data, scores and retention beyond plan allowances |
| Braintrust Enterprise self-hosting contract | Braintrust | Recurring | Contract scope; the control plane stays with Braintrust |
| Langfuse software license | Langfuse (only if you buy Enterprise) | Recurring if Enterprise | Core features are MIT licensed; RBAC, audit logs, SCIM and retention need an Enterprise key |
| Compute for Langfuse Web, Worker and PostgreSQL | Your cloud provider | Recurring | Trace volume and worker scaling on CPU load; minimums are 2 CPU and 4 GiB each |
| ClickHouse, Redis or Valkey and blob storage | Your cloud provider | Recurring | ClickHouse needs at least 2 CPU and 8 GiB, with 16 GiB memory for larger deployments; storage grows with retained traces |
| Implementation and data migration | Your team or an implementation partner | One-time | Re-instrumenting applications, exporting Braintrust data and mapping scores and datasets |
| Upgrades and major version moves | Your team or a support retainer | Recurring, at each major version | v3 to v4 requires ClickHouse 25.12 or later, PostgreSQL 15 or later and Redis 7.0 or later, plus about 3x ClickHouse disk headroom for backfill |
| Internal skills, on-call and support | Your staff, plus optional Enterprise support | Recurring | Who operates ClickHouse, backups and storage; Enterprise adds a named support engineer and SLA at custom pricing |
What moves the cost comparison most
Braintrust's bill grows with usage and features, while self-hosted Langfuse moves cost into infrastructure and people. The drivers below decide which side of the comparison gets more expensive as your usage, teams and compliance needs grow.
Trace and score volume
Braintrust charges for processed data and scores, so higher trace and evaluation volume raises the managed bill directly. Self-hosted Langfuse spreads that load across ClickHouse, the worker containers and object storage, so volume raises compute and storage sizing instead of a per-unit fee. Ask Braintrust for the processed-data and score overage rates in writing, then estimate your monthly volume from a representative week of production traces.
Retention period
Longer retention increases storage on the Braintrust side through overages and, on the Langfuse side, through ClickHouse disk and blob storage that you size and back up yourself. Retention policies in self-hosted Langfuse are an Enterprise license key feature, so a longer retention need can also change the license decision. Confirm the retention period your compliance team requires before comparing totals.
Number of environments and teams
Each extra environment, such as staging, pre-production and production, means another Langfuse deployment to provision, patch and back up, while Braintrust usually bills by usage across projects. Shared teams can share one self-hosted instance with project-level roles, but project-level RBAC requires an Enterprise key. Count environments and teams early, because they multiply infrastructure and operating work.
Availability requirements
Docker Compose on a single VM is documented for testing and has no high availability, so it suits evaluation rather than production tracing. Production deployments follow the Kubernetes with Helm path, which needs cluster capacity, a redundant PostgreSQL, ClickHouse and Redis or Valkey layer, and on-call coverage. Higher availability targets are the single biggest reason self-hosting costs rise, so define the uptime you need before choosing the architecture.
Enterprise features you need
SSO, audit logs, data masking, SCIM provisioning and data retention policies in self-hosted Langfuse require an Enterprise license key, and the Enterprise plan adds a support SLA. Braintrust's Enterprise tier is priced by contract. Make a list of the controls your security review requires, then ask both vendors which tier provides each one before comparing totals.
Skills already in house
Running ClickHouse, PostgreSQL, Redis or Valkey and blob storage requires people who can upgrade them, tune them and restore them from backup. The Langfuse v3 to v4 guide requires ClickHouse 25.12 or later and PostgreSQL 15 or later, and it advises about 3x ClickHouse disk headroom when backfilling historic data. Teams with Kubernetes and database operators carry this cost at lower marginal effort, while teams without them should price in outside support or a retainer.
How to build the comparison with your own numbers
This sequence turns the cost components into a dated, written comparison of Braintrust and self-hosted Langfuse in your own environment, so the annual totals rest on invoices, quotes and sizing you can defend.
Collect twelve months of Braintrust invoices
Finance or procurement pulls every Braintrust invoice for the last twelve months, separating the base plan fee, processed-data charges, score volume, and retention overages. Ask Braintrust in writing to explain any line item you cannot trace to usage. The step is done when you have a monthly table of each cost line and a clear view of which months spiked and why.
Request a written Enterprise quote for self-hosting
Ask Braintrust for a written quote covering the self-hosted data plane scope, including what the control plane still runs in Braintrust's environment, support terms, and any minimums. Confirm which elements you would still pay for if you self-host. The step is done when you hold a written scope and price that you can place next to the other lines in the model.
Size the Langfuse deployment from documented minimums
Your platform engineers size the Langfuse deployment using the minimum resources in Langfuse's scaling guide for Langfuse Web, Langfuse Worker, PostgreSQL, Redis or Valkey, and ClickHouse, then add headroom for your trace volume. Note that the guide recommends at least 16 GiB of ClickHouse memory for larger deployments, and that a single-VM Docker Compose setup has no high availability. The step is done when you have a resource table for test and production environments.
Price the cloud infrastructure in your own accounts
Your cloud or infrastructure team prices the compute, managed or self-run PostgreSQL, ClickHouse nodes, Redis or Valkey, object storage, backups, and network egress from your own cloud or data center contracts. Keep production and non-production environments as separate lines. The step is done when each sized component has a monthly and annual cost from your own price sheet, with the recurring items clearly marked.
Estimate engineering time for operations and upgrades
Engineering managers estimate the recurring hours for operating the stack, applying minor updates, planning major upgrades such as the documented v3 to v4 path, monitoring ClickHouse disk growth, and on-call coverage. Use your own on-call rota and past incident data rather than a generic figure. The step is done when each recurring activity has an owner, a frequency, and an hourly cost from your internal rates.
Add implementation and support lines from written scopes
Request written scopes and quotes from any external partner for initial implementation, hardening, backup design, and ongoing support, and from Langfuse if you consider its Self-Hosted Enterprise plan for a named support engineer and SLA. Compare each line with its deliverables so you are not pricing unlike work. The step is done when every external cost has a written scope, a deliverable list, and a start-and-end definition.
Compare annual totals and volume sensitivity
Build a single annual model with a one-time column for implementation and migration effort and a recurring column for subscriptions, infrastructure, and staff time. Then test the model at your current volume and at higher volumes for traces, scores, and retention, because processed-data and retention overages scale with usage while self-hosted infrastructure scales in steps. The step is done when you can show the annual totals for both options at several volume levels and name the break-even point, if one exists.
What the Langfuse license covers and where Enterprise keys apply
The Langfuse repository is MIT licensed everywhere except the enterprise directories: the ee, web/src/ee and worker/src/ee paths fall under a separate enterprise license. Langfuse states that all core features and APIs are available in the MIT-licensed open source version without limits, and that a self-hosted deployment runs the same codebase as Langfuse Cloud. For a cost model, this means tracing, datasets, evaluations and prompt management do not need a license key. The question is not whether the core platform is usable, but which governance controls your security or compliance reviewers will require on top of it.
Several controls need an Enterprise license key on a self-hosted instance: project-level RBAC roles, protected prompt labels, data retention policies, audit logs, server-side data masking, UI customization, organization creators, the org management API with SCIM, and the instance management API. Before you price the migration, list which of these your team depends on today, then check the Braintrust contract for each one, since access controls, retention settings and audit trails are often bundled into higher Braintrust tiers. Any control your Braintrust agreement covers and Langfuse's open-source version does not will need either a Langfuse Enterprise key or an internal workaround, so record that gap in the comparison. Confirm the exact feature list with Langfuse, because the license key documentation is the authority on which features are gated.
The Self-Hosted Enterprise plan is custom priced, and the Langfuse self-hosted pricing page describes what it adds beyond the free open source plan, including a named support engineer, solutions architect access and a support SLA. Ask for a written quote that names each gated feature you need, the support level, and whether the key term renews annually. Compare that quote with your Braintrust renewal, not with the open-source plan alone, because the open-source plan does not include the same vendor support. Where you choose the open-source plan, budget internal or partner effort for the controls that the key would otherwise provide.
Questions buyers ask before comparing costs
How much does a self-hosted Langfuse deployment cost?
The software is free under the MIT license for core features, so the cost is the infrastructure and the people who run it. Your bill comes from compute for the Langfuse Web and Worker containers, PostgreSQL, Redis or Valkey, ClickHouse and S3-compatible object storage, plus the engineering time to operate them. Langfuse's scaling guide lists minimum resources of 2 CPU and 4 GiB for web, worker and PostgreSQL, 1 CPU and 1.5 GiB for Redis or Valkey, and 2 CPU and 8 GiB for ClickHouse, with at least 16 GiB of ClickHouse memory for larger deployments. Cost rises with trace volume and retention. The Self-Hosted Enterprise plan is custom priced and sold separately, so ask for its quote only if you need features behind a license key.
How long does a self-hosted Langfuse setup take?
Setup time depends on your cloud environment, your network and security controls, how many applications send traces, and whether you need high availability. A Docker Compose test on a single VM is quick to start but is not highly available, so it is not a production design. Production deployments on Kubernetes with Helm or Terraform on AWS, Azure or GCP need more planning. Ask for a timeline in a written, fixed scope that lists the environments, components, integrations and handover deliverables, rather than relying on a general estimate.
Can a small team run self-hosted Langfuse on its own?
A small team can run it, but it should count the components before deciding. A production setup involves two application containers, Langfuse Web and Langfuse Worker, plus PostgreSQL, ClickHouse, Redis or Valkey and S3-compatible storage. Each needs backups, patching, monitoring, capacity planning and access control. Operating ClickHouse in particular requires database skills that many application teams do not have. Check whether your platform or SRE team already runs PostgreSQL and Kubernetes, and whether anyone can own ClickHouse disk growth and upgrades before you commit.
What does the work need from our team?
You need a named owner for the platform, access to your cloud accounts or data center, a decision on where traces and prompts may be stored, and people who can review security settings and approve changes. You also need application engineers to update instrumentation so that traces reach Langfuse. Ask any implementation partner to list these inputs in the written scope so you know your share of the work before it starts. DeployOpen plans, implements, integrates, hardens, backs up and documents self-hosted platforms inside your own infrastructure, and most engagements are fixed-scope.
How do upgrades affect effort and cost?
Updates within a major version are non-disruptive, and database migrations apply automatically at startup. Major upgrades follow separate migration guides, so they take more planning. For example, the v3 to v4 guide requires ClickHouse 25.12 or later (26.4 recommended), PostgreSQL 15 or later and Redis 7.0 or later, moves to an observations-first ClickHouse data model, and advises about 3x ClickHouse disk headroom when backfilling historic data. Budget for test environments, storage headroom and rollback planning each time a major version is released.
What happens to the license key during a major version upgrade?
Features that need an Enterprise license key, such as project-level RBAC roles, data retention policies, audit logs, server-side data masking and SCIM provisioning, are tied to a key that you hold for your instance. Langfuse's license key documentation describes how the key is applied, and you should confirm with Langfuse, in writing, that your current key and feature set remain valid on the target major version before you start the upgrade. Include this check in your upgrade test plan so a key problem does not surface in production.
Do I still need a support contract if I run Langfuse myself?
The open-source version has no vendor support plan, so your team carries the operational responsibility. Langfuse's Self-Hosted Enterprise plan adds a named support engineer, solutions architect access and a support SLA, and its terms are set individually. Ongoing support, upgrades and continuous testing can also be covered by a DeployOpen retainer. Decide which of these your team can absorb internally and which it wants under contract, and compare the quotes against the same scope of work.
What migration adds to the first-year total
Migration is a separate line in the first-year total, and it is easy to leave out of a Braintrust versus Langfuse comparison because the subscription numbers look complete on their own. The work falls into four parts: exporting logs, datasets and experiments from Braintrust, mapping scores and metadata onto Langfuse's data model, re-instrumenting applications so traces land in Langfuse, and rebuilding evaluators and CI gates that currently run against Braintrust. Most of this effort is paid as internal engineering time, or as partner time if you scope it with DeployOpen, and it is one-time. Ask any implementation partner to price each of the four parts separately so you can see which ones your own codebase makes expensive.
Running both platforms in parallel is the cost that people underestimate. While Braintrust remains your system of record and Langfuse is validated, you pay for two sets of infrastructure and two sets of operators, and engineers maintain two instrumentation paths. Parallel running is one-time in intent but recurs in practice for as long as evaluation results disagree. Define the exit condition before you start, such as agreed scores on a fixed evaluation set and a sign-off from the team that owns the CI gates, so the parallel period has an end date. Confirm the retention period you need for Braintrust history, because that determines how long the old project must stay readable and paid for.
Keep the migration estimate separate from the steady-state run cost, and check that the quote covers the ClickHouse and PostgreSQL capacity needed during backfill, since historic data imports consume storage and compute before the new system is in service. The step-by-step export procedure, including Braintrust's export options and the paging approach for large projects, is covered on the Braintrust data export guide, and the sibling migration page sets out the sequence for moving to Langfuse. Use those pages to build the work plan, then price it against the four components above.
How DeployOpen scopes a fixed-price Langfuse implementation
DeployOpen is a Chainscore Labs company that plans, implements, hardens, backs up and documents self-hosted open-source platforms inside your own cloud accounts, data centers or air-gapped networks. For a Braintrust to self-hosted Langfuse move, we start from a written scope that names every component the cost comparison depends on.
How we scope
- We confirm which Braintrust features your teams use, such as tracing, scorers, datasets and prompt management, before anything is sized.
- We agree in writing which Langfuse features need an Enterprise license key and whether your requirements depend on them.
- We document the target ClickHouse, PostgreSQL, Redis or Valkey and object storage layout, plus who on your side will operate them.
- Engagements are fixed-scope, so deliverables and timelines are agreed in writing before work starts, and a quote follows from that scope.
How we build
- We deploy Langfuse Web and Langfuse Worker with Helm or Terraform into the Kubernetes environment you approve.
- We size the ClickHouse, PostgreSQL and storage tiers against your expected trace volume and retention settings, and record the assumptions.
- We wire SSO, network isolation, backups and secrets handling to your existing identity and security controls.
- We test ingestion, evaluation runs and restore procedures before any production traffic moves to the new platform.
How we hand over
- You receive architecture diagrams, configuration records, runbooks for upgrades, backups and scaling decisions.
- We document the cost components and what recurs, so your finance and platform teams can track the running cost.
- We provide evidence of configuration and controls that your internal or external auditors can review, without issuing any audit opinion ourselves.
- We walk your operators through the handover so the team can run the platform without depending on us day to day.
How we support
- Ongoing support, upgrades and continuous testing are available as retainers after handover.
- Major Langfuse upgrades follow dedicated migration guides, and we plan and test each one in a staging environment first.
- Tell us your reporting, change control and on-call expectations so the retainer scope matches how your team works.
