Where self-hosted Langfuse costs less than Arize, and where it costs more
A renewal or steady growth in span volume is usually what forces a cost review of Arize AX. The Langfuse software is MIT licensed outside its ee directories, so the recurring cost moves from a vendor subscription to your infrastructure, operations, upgrades and, if you need them, the optional enterprise license key and support. Whether that is cheaper depends on three things you control: how many traces you ingest, how long you keep them, and whether your team can run PostgreSQL, ClickHouse, Redis or Valkey, and object storage well. Before comparing, obtain Arize's plan terms, span allowances, retention periods and overage rules in writing for your specific contract, because those figures are not part of this comparison and vary by agreement.
On the self-hosted side, the cost components are the cloud or data center resources for the two Langfuse application containers, Langfuse Web and Langfuse Worker, plus PostgreSQL, ClickHouse, Redis or Valkey and S3-compatible storage. Langfuse's scaling guide lists minimum sizes 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. ClickHouse storage is the main variable: retention length and ingestion volume set the disk you buy and the headroom you keep. Labor is the other recurring line. Someone must own Kubernetes or Docker Compose operations, backups, monitoring and security patching, and that effort is largely fixed regardless of trace volume. Ask your team or DeployOpen to estimate it against your own environment, since no effort figure is supplied here.
Plan for upgrades as a recurring and sometimes expensive task. Updates within a major version apply database migrations at startup, but the Langfuse v3 to v4 guide requires ClickHouse 25.12 or later, PostgreSQL 15 or later and Redis 7.0 or later, moves to an observations-first data model, and advises about 3x ClickHouse disk headroom when backfilling historic data. Budget that headroom before you commit to long retention. The enterprise key, which unlocks features such as data retention policies, audit logs, SCIM and the instance management API, is priced by Langfuse on a custom basis, so request it as a separate line item with its support scope. Ask Arize to confirm in writing which features and support levels your current contract includes, so the comparison is like for like.
Cost components on each side and whether they are one-time or recurring
This table sets each cost component of a commercial Arize AX subscription beside the matching components of a self-hosted Langfuse deployment, and marks whether each is one-time or recurring. Arize pricing is set by plan and contract, so confirm span allowances and retention terms in writing before comparing totals. Most of the difference sits in who runs the infrastructure and who carries the upgrades, not in the open-source software itself.
| Cost component | Commercial Arize AX side | Self-hosted Langfuse side | One-time or recurring |
|---|---|---|---|
| Software license or subscription | Enterprise or plan subscription, priced by contract. Ask Arize for the written plan terms. | MIT-licensed core with no license fee. Enterprise license key and Self-Hosted Enterprise plan are custom-priced and quoted separately. | Arize: recurring. Langfuse core: none. Enterprise key: recurring if chosen |
| Span or usage charges | Billed against plan span allowances. Confirm overage rules with Arize. | No span metering in the open-source version. Cost shifts to storage and compute you size yourself. | Arize: recurring and usage-driven. Langfuse: indirect, via infrastructure |
| Retention tier | Trace retention depends on the plan you buy. Confirm the retention window for your plan. | Retention is set in your ClickHouse storage capacity. Data retention policies are an Enterprise license key feature. | Arize: recurring. Langfuse: storage recurring, policy feature requires key |
| ClickHouse | Not applicable, handled by Arize. | 2 CPU and 8 GiB minimum, at least 16 GiB memory for larger deployments, plus disk headroom for backfills. | Recurring cloud or hardware cost, with one-time sizing |
| PostgreSQL, Redis or Valkey, object storage | Not applicable, handled by Arize. | PostgreSQL 4 GiB minimum with 2 CPU, Redis or Valkey 1 CPU and 1.5 GiB, plus S3-compatible blob storage. | Recurring cloud or hardware cost |
| Compute for web and worker | Not applicable, handled by Arize. | Web and worker containers scale on CPU load and ingestion volume, with 2 CPU and 4 GiB minimum each. | Recurring |
| Implementation and migration | Re-instrumentation, exporter changes and dataset rebuilds, often done in-house. Ask for vendor services if needed. | Helm or Terraform deployment, OpenTelemetry repointing, dataset and evaluator rebuilds. Scope and quote separately. | One-time, but effort varies with trace volume and evaluator count |
| Support | Included in the plan tier you buy. Confirm the response terms in your contract. | Community support on the open-source version. Self-Hosted Enterprise adds a named support engineer and support SLA, quoted separately. | Arize: recurring. Langfuse: recurring if an enterprise plan is chosen |
| Upgrades | Handled by Arize on the managed service. | Minor updates apply migrations at startup. Major upgrades, such as v3 to v4, need planned downtime or a migration project and extra ClickHouse headroom. | Recurring internal effort, with periodic one-time migration projects |
| Internal skills | Limited to instrumentation and evaluation work. | Kubernetes, PostgreSQL, ClickHouse and object storage operations, plus backups and monitoring. Staff or contract these skills. | Recurring, usually the largest line for teams without platform staff |
What moves the cost of each side up or down
On the Arize side, cost follows span allowances, retention and plan tier. On the self-hosted Langfuse side, cost follows the infrastructure you size, the edition you need and the people who run it. The cards below tie each driver to the line it moves, so you can ask the right questions before comparing totals.
Trace and span volume
Span volume is the main input on the Arize AX side, because plan allowances are counted in spans. On the self-hosted side, the same volume drives ClickHouse storage, write load and worker CPU, so the cost moves with disk and compute instead of a subscription line. Ask Arize for the span allowance and overage terms in your plan, then estimate your own daily span count from a representative week of production traffic before you size anything.
Retention period
Retention is where the two models diverge most. Arize AX retention depends on the plan you buy, so keeping data longer usually means a higher tier. In a self-hosted Langfuse deployment, retention is a storage decision: every extra month of traces adds ClickHouse disk, and Langfuse's data retention policies are an Enterprise license key feature. Confirm the retention you need for audit and debugging, then price the disk for that window.
Projects and users
Project count and user count affect different lines. Seats and project access are typically part of a commercial contract, which you should confirm with Arize in writing. For self-hosted Langfuse, the open-source edition has no user or project limits, but project-level RBAC roles, SCIM and the org management API need an Enterprise license key. Map which teams need isolated projects and which need only read access before you decide on the edition.
Edition and enterprise key features
The edition decides whether you pay a license fee at all. Langfuse's core features run in the MIT-licensed open-source version, while RBAC, audit logs, server-side data masking and SCIM need an enterprise license key with a custom-priced plan. Compare those features against what your Arize contract already includes, such as SSO and support terms, and ask Langfuse which of your compliance controls depend on a key before you assume the open-source edition covers them.
Availability target
A single VM running Docker Compose is enough for testing, but Langfuse documents it as having no high availability. Production runs on Kubernetes with Helm or Terraform, which adds cluster, load balancer, managed database and backup costs, and the operating effort that comes with them. Decide the availability target first, because it sets the number of nodes, replicas and managed services you pay for every month, and it is the largest recurring infrastructure line.
In-house skills and upgrade effort
Self-hosting moves cost from the vendor invoice to your payroll or a partner's scope. Running PostgreSQL, ClickHouse, Redis or Valkey and S3-compatible storage requires skills in each, plus a plan for upgrades. Langfuse applies minor-version migrations at startup, but major upgrades such as v3 to v4 need a dedicated guide and extra ClickHouse disk headroom for backfills. Count the platform engineers who can own each component, and decide whether you will staff this in-house or buy a fixed-scope implementation and support retainer.
What a self-hosted Langfuse deployment runs and what it needs
A self-hosted Langfuse deployment is six moving parts, not one application. Two application containers do the work: Langfuse Web serves the UI and APIs, and Langfuse Worker processes events asynchronously after ingestion. Around them sit PostgreSQL, ClickHouse, Redis or Valkey, and S3-compatible blob storage. Each of these is a service your team will patch, back up, monitor and pay for, so the infrastructure budget should be built from this list rather than from the software license line. The Langfuse 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. Treat these as floors for a working install, not as sizing advice.
Size the deployment from your own trace volume, not from the minimums. ClickHouse usually carries most of the storage and query load, so the number that drives your cost is how many observations you ingest per day and how long you retain them. Ask your team to measure current Arize span volume, the retention window you need, and the peak ingestion rate during batch evaluation runs. Then plan the Worker count against CPU load: Langfuse recommends scaling worker containers on CPU, and under heavy ingestion it suggests splitting langfuse-web into separate ingestion and user interface deployments. Those splits add nodes and load balancer rules, so they belong in the quote as an explicit scope item.
Before you commit to a topology, confirm the Langfuse version you will run and the matching ClickHouse, PostgreSQL and Redis versions in the upgrade guide for that release, because the v3 to v4 migration sets its own minimums and asks for extra ClickHouse disk during backfill. Choose between Docker Compose on a single VM, which Langfuse documents for testing without high availability, and the Kubernetes Helm or Terraform deployments it documents for production on AWS, Azure and GCP. The production path is where the recurring operational work sits: backups for each stateful service, a tested restore procedure, and an upgrade calendar.
How to build the Arize versus Langfuse comparison with your own numbers
This sequence turns a vague sense that Arize is expensive into a written, side-by-side comparison of current Arize charges and self-hosted Langfuse costs over three years, built from figures your vendors and your own infrastructure team can confirm.
Export current span volume and retention needs
The platform owner pulls span counts for the last several months from Arize AX usage views, then records peak days, average trace size and the retention window your compliance, debugging and evaluation teams actually need. Separate production traffic from development and evaluation runs, because they often differ sharply. Done when you have a monthly span baseline, a growth assumption agreed with the product owner and a written retention requirement per data class. Ask Arize to confirm how each figure is metered so your numbers match its billing definitions.
Request written plan terms and overage rules
Procurement asks Arize for a written quote covering the plan tier, included span allowance, retention period for each tier, overage rates, minimum commitments, renewal uplift terms and the support level included. Ask whether data residency, SSO and any self-hosted AX option are included or sold separately. Done when you hold a dated quote that states each limit in the vendor's own words, and you can point to the clause that governs overage charges.
List Langfuse components and map them to your prices
The infrastructure team lists every component of a self-hosted Langfuse deployment: the Web and Worker containers, PostgreSQL, ClickHouse, Redis or Valkey, and S3-compatible blob storage, plus the load balancer, DNS, certificates and monitoring. For each component, record the instance or managed-service size using the minimum resources in Langfuse's scaling guide as a floor, then price it at your cloud provider or data center rates. Done when each line has a unit price, a quantity and a source document, and the sheet shows which components you would run yourself and which you would buy as managed services.
Add engineering time for installation and operations
Platform engineering estimates the internal effort for initial installation on Kubernetes with Helm or Terraform, configuring backups for PostgreSQL, ClickHouse and object storage, setting up monitoring and on-call coverage, and performing upgrades. Upgrades within a major version apply migrations at startup, while major upgrades follow dedicated migration guides and need planning time. Record the people involved and the hours they spend, using your own loaded rates. Done when the engineering lines show one-time setup separately from recurring operations, and each assumption is written down so reviewers can challenge it.
Decide on the enterprise key and support need
Compare the open-source feature set with your requirements. Some capabilities, including project-level RBAC, data retention policies, audit logs, server-side data masking and SCIM provisioning, need an Enterprise license key. Langfuse's self-hosted pricing page describes a custom-priced Self-Hosted Enterprise plan with a named support engineer and a support SLA. Decide whether your team can carry support in-house or needs a contract. Done when you have a yes or no for each required feature, a support decision, and a request for written pricing for the enterprise key if you need it.
Build a three-year view with separated cost lines
Create a spreadsheet with one tab per option. Put one-time costs, such as initial setup, migration of traces and evaluators, and training, in their own rows. Put recurring costs, such as Arize subscription fees, overage charges, cloud infrastructure, storage growth, enterprise key fees and staff time, in separate rows by year. Model the ClickHouse storage growth from your retention decision, and note the extra disk headroom Langfuse's v3 to v4 guide advises for backfilling historic data. Done when both options show totals for each of the three years and every number traces to a named source.
Confirm figures in writing with each vendor
Send the completed model to Arize and ask for written confirmation of the plan terms, span allowance, retention and overage figures you used. Ask Langfuse to confirm the enterprise key and support pricing if you plan to use them. Ask your cloud provider or data center team to confirm the infrastructure rates. Done when each figure in the model is either confirmed in writing by its source or marked as an assumption, and the final comparison is reviewed by finance, security and platform engineering together.
How the license of each self-hosted option changes the cost model
Langfuse is open core. Code outside its enterprise directories is MIT licensed, and the project states that all core features and APIs are available in the open source version without limits. A self-hosted deployment runs the same codebase as Langfuse Cloud, so you do not pay a license fee to run the core product. Your cost is the infrastructure and the people who run it: two application containers, Langfuse Web and Langfuse Worker, plus PostgreSQL, ClickHouse, Redis or Valkey, and S3-compatible blob storage.
- No usage fee for the core product: Trace volume is not metered by the license, so cost grows with ClickHouse storage, compute and retention you configure.
- Enterprise key for specific features: Project-level RBAC, data retention policies, audit logs, server-side masking and SCIM need an Enterprise license key.
- Paid support is optional: The self-hosted pricing page describes a custom-priced Self-Hosted Enterprise plan with a named support engineer and an SLA.
- Recurring work: Upgrades, backups and capacity planning fall to your team or to a support retainer.
- Enterprise directories are separate: Code under the ee directories has its own license, so confirm which features you rely on.
Before budgeting, confirm with Langfuse which features on your list need the key and ask for the Self-Hosted Enterprise terms in writing.
Arize AX is a managed enterprise platform, and it also offers a self-hosted option that runs on Kubernetes in GCP, AWS, Azure, IBM Cloud, OpenShift or Rancher environments. The self-hosted edition is a commercial arrangement, so the license and support terms come from your contract with Arize rather than from an open-source license. Your bill therefore covers the subscription terms plus the infrastructure you run underneath.
- Contracted license: Pricing, term and support scope are set in the agreement, not in a public license file.
- Infrastructure still yours: The Kubernetes cluster, storage and networking are your cost, as with any self-hosted stack.
- Operations still yours: Patching, scaling and upgrades remain with your platform team unless the contract says otherwise.
- Renewal leverage: Because the price is negotiated, the cost moves with seat, volume and support decisions at each renewal.
- Ask for these items: Written scope for what the self-hosted edition includes, upgrade support, and the notice period for price changes.
Confirm directly with Arize which features are included in the self-hosted edition, since the managed and self-hosted feature sets can differ.
Phoenix is Arize's open-source tracing and evaluation tool, built on OpenTelemetry and OpenInference. It is licensed under the Elastic License 2.0, which is not an OSI-approved open-source license. For internal self-hosting there is no usage fee from the license, so the direct software cost is zero, and your spend is the infrastructure and staff time to run it.
- No license fee for internal use: Running Phoenix in your own environment carries no license charge.
- No usage metering: Phoenix does not impose the span allowances that AX plans use.
- Managed service limit: The license forbids offering Phoenix to third parties as a hosted or managed service, which matters if you resell or host it for customers.
- Feature scope: Compare what Phoenix covers with the AX features your teams use before assuming parity.
- Your own operations: Storage, backups and upgrades are on your team, and they grow with trace volume.
Have legal or procurement review the ELv2 terms against your intended use, especially if any part of the deployment serves outside parties.
Opik, built by Comet, is licensed under Apache-2.0. Its README states that the full platform, including the server backend, web application, tracing, evaluations and prompt management, can be self-hosted without a commercial license. That makes it the option in this comparison where the license imposes no fee and no hosted-service restriction on the platform itself.
- No commercial license required: The full platform is self-hostable under Apache-2.0 as stated in the project README.
- Storage sized by you: Opik uses ClickHouse-based storage, so retention and capacity are set by your own sizing rather than by a plan's span limit.
- Infrastructure and staff are the costs: Compute, ClickHouse disks, backups and on-call time recur each month.
- Support is separate: Any paid support comes from a vendor or partner agreement you negotiate, and its scope should be written down.
- Growth drivers: Trace volume, retention length and evaluation workload push infrastructure cost up.
Size ClickHouse against your expected trace volume and retention window before comparing totals with a span-based plan.
What upgrades and support add to the recurring cost
Upgrades are the recurring cost most often underestimated in a self-hosted Langfuse deployment. Langfuse states that updates within a major version are non-disruptive and that database migrations run automatically at startup, so a minor release is mostly a matter of pulling new images, checking the rollout and confirming traces still arrive. Major versions are different. The v2 to v3 and v3 to v4 changes each follow a dedicated migration guide, and the Langfuse upgrade guide is the starting point for planning them. Budget staff time for reading the guide, staging the upgrade and validating dashboards, evaluators and API integrations before production cutover. The v3 to v4 guide raises the floor on your infrastructure. It requires ClickHouse 25.12 or later, with 26.4 recommended, PostgreSQL 15 or later and Redis 7.0 or later. It also moves to an observations-first ClickHouse data model and advises about 3x ClickHouse disk headroom when backfilling historic data. Check the current versions of each component in your environment before scheduling the work, because a database upgrade may be needed first. Ask whether your ClickHouse capacity covers the backfill window, and confirm the exact disk figure in the guide rather than relying on this summary. The Self-Hosted Enterprise plan is the commercial option for organizations that want outside help with this work. Langfuse's self-hosted pricing page describes it as a custom-priced plan that adds a named support engineer, solutions architect access and a support SLA, on top of the free open-source plan. Ask Langfuse what the SLA covers, what response times apply and whether the license key is included, and ask for the price in writing. Effort for upgrades is recurring and depends on release cadence and your internal skill with Kubernetes, Helm and ClickHouse, so a team with thin database experience should plan for either a retainer or a support contract.
Questions buyers ask about Arize and self-hosted Langfuse costs
How much effort does moving from Arize to self-hosted Langfuse take?
The effort depends mainly on how many services you run, how much trace history you migrate, and how many evaluators and datasets must be rebuilt. Self-hosted Langfuse runs two application containers, Langfuse Web and Langfuse Worker, plus PostgreSQL, ClickHouse, Redis or Valkey, and S3-compatible blob storage. Each of those is a component your platform team must provision, patch, back up and monitor. Your Arize exporters also need repointing to Langfuse, which is usually small work for OpenTelemetry-based instrumentation but grows with the number of applications. Ask any implementation partner to split its quote by component, data migration and evaluator rebuild so you can see where the effort sits.
Can our team run self-hosted Langfuse without outside help?
Yes, many platform teams run it in-house, but the answer depends on your Kubernetes, database and storage skills. Langfuse documents Docker Compose on a single VM for testing, which it notes has no high availability, and Kubernetes with Helm or Terraform for production on AWS, Azure and GCP. Production operation means tuning ClickHouse, handling major upgrades such as the v3 to v4 path, and scaling Langfuse Web and Worker under load. Assess your team's experience with ClickHouse specifically, since it is less common than PostgreSQL. Outside help can be scoped as a one-time build, an ongoing retainer for upgrades and testing, or a mix of both.
What does the Langfuse Enterprise license key add, and when is it needed?
The Enterprise license key unlocks a specific set of features, not the core tracing and evaluation product. Langfuse states that all core features and APIs are available in the MIT-licensed open source version without limits. Features that need the key include project-level RBAC roles, protected prompt labels, data retention policies, audit logs, server-side data masking, UI customization, and the org management API with SCIM. You need the key only if your governance requirements call for these controls. Confirm with Langfuse which of your requirements map to the key, and whether the Self-Hosted Enterprise plan on its pricing page, which adds a named support engineer and a support SLA, is needed or optional.
How does span volume change the infrastructure we need?
Span volume drives ClickHouse storage and compute more than any other component, and it also shapes how many Worker and Web containers you need. Langfuse recommends scaling worker containers on CPU load and, under heavy ingestion, splitting langfuse-web into separate ingestion and user interface deployments. Langfuse's scaling guide lists minimum resources of 2 CPU and 8 GiB for ClickHouse, with at least 16 GiB of ClickHouse memory for larger deployments. Your retention period multiplies storage needs, which is the main difference from Arize's span allowances by plan. Estimate expected spans per day and retention length first, then test against your own workload before fixing disk and memory sizes.
How long does a migration from Arize take, and what does it depend on?
Timing depends on the scope you agree, not on a fixed number that applies to every team. The main variables are the number of applications being re-instrumented, the volume of historical data you need to bring across, the count of datasets and evaluators to recreate, and the production change windows you must respect. Langfuse's v3 to v4 upgrade guide advises about 3x ClickHouse disk headroom when backfilling historic data, which affects both storage planning and the sequence of work. Ask for a timeline in the written scope that lists each phase and the decisions it depends on, and confirm whether Arize data export will be run by your team or by the partner.
What should we ask for in a fixed-scope quote for this migration?
Ask for a written scope that names each deliverable, its acceptance criteria and who owns each step. Most engagements are fixed-scope, with scope, deliverables and timelines agreed in writing before work starts. A useful quote separates one-time work, such as architecture, deployment, migration and hardening, from recurring work, such as upgrades, backups and testing retainers. It should state which environments are covered, how cloud or data center infrastructure is provisioned and who pays for it, and what documentation and evidence you receive. Request a line for ClickHouse sizing and load testing, and for any Enterprise key or support tier you choose to include.
Does self-hosting Langfuse change our compliance position?
Self-hosting keeps trace data in infrastructure you control, but it does not make you compliant by itself. DeployOpen prepares systems, controls and evidence for compliance assessments, while independent auditors and certification bodies issue any audit reports or certifications. Your assessor will look at access control, backup, retention, encryption and logging for every component, including ClickHouse, PostgreSQL, Redis or Valkey and object storage. Note that data retention policies and audit logs are among the features that need the Enterprise license key, so confirm early whether your controls depend on them.
How DeployOpen scopes a fixed-scope Langfuse cost and implementation plan
DeployOpen is a Chainscore Labs company that plans, implements, hardens, backs up, upgrades and documents self-hosted open-source platforms inside your own infrastructure. For Arize cost vs self-hosted Langfuse, we start from your current Arize AX spend, retention needs and data location, then write a fixed scope before any work begins.
How we scope
- Inventory current Arize AX span volumes, plan tier, retention needs and the data residency rules you must meet.
- Size the Langfuse components (web, worker, PostgreSQL, ClickHouse, Redis or Valkey, object storage) against your ingestion profile.
- Agree deliverables, timelines and responsibilities in writing, so the quote follows directly from the written scope.
How we build
- Deploy a Kubernetes and Helm reference environment in your cloud account, data center or air-gapped network.
- Configure SSO, backups and storage retention, and decide with you which enterprise key features your team actually needs.
- Test ingestion throughput, query performance and failure recovery, then record the results against the agreed scope.
How we hand over
- Document the architecture, sizing decisions, upgrade path and the ClickHouse disk headroom needed for historic backfills.
- Provide runbooks for upgrades, including the v3 to v4 migration requirements, so your team can repeat them safely.
- Deliver the handover package with the configuration and evidence your internal auditors or assessors may ask to review.
How we support
- Engagements are fixed-scope, and ongoing support, upgrades and continuous testing are available as a retainer.
- Retainers can cover Langfuse version upgrades, backup restore tests and capacity reviews as ingestion grows.
- DeployOpen prepares systems and evidence for compliance assessments but does not issue audit reports or certifications.
