When Langfuse Cloud fits and when self-hosting fits
Choose Langfuse Cloud when your traces can leave your network under your contract terms and your team does not want to run databases. Choose self-hosted Langfuse when traces must stay inside infrastructure you control, or when you already run PostgreSQL, ClickHouse, Redis or Valkey and object storage well enough to add one more stack. The MIT edition of self-hosted Langfuse includes all core features without limits, so the decision is mostly about where the data may sit, who operates the stack and which paid tiers you need, not about feature access. The first condition is data location. Traces often carry prompts, completions, retrieved documents and user identifiers, so a team handling customer data should start by asking legal and security what may be stored where. Langfuse Cloud offers US, EU or JP data regions, and its pricing page places a HIPAA-ready region and SOC 2 and ISO 27001 reports in the Pro tier. Confirm with Langfuse which region and report apply to your data before you send production traces, and confirm your own residency obligations with counsel.
The second condition is operational capacity. A self-hosted deployment runs a web and a worker container plus PostgreSQL, ClickHouse, Redis or Valkey and S3-compatible storage, and Langfuse's scaling guide lists minimum sizes for each. Someone must patch, back up, monitor and size those services, and handle major upgrades such as the v3 to v4 move, which requires ClickHouse 25.12 or later. Those costs are recurring and fall on your staff or a partner you pay under a written scope. Cloud shifts that work to Langfuse but bills by unit, so estimate your traces, observations and scores per month from the billable units definition before you assume the cheapest tier. As a practical test, pick the region and compliance evidence you need, then ask whether your team can meet the Langfuse Docker Compose or Kubernetes baseline with monitored backups and a tested upgrade path. If the answer to either question is no, cloud is the likelier fit. If both are yes and your residency rules require your own infrastructure, self-hosting is the right choice even though you carry the operations.
How the two models compare on license, features and data control
This table compares Langfuse Cloud and self-hosted Langfuse on license, data location, regions, compliance reports, SSO, data access windows, and cost structure. Read each row as a condition to check against your own data rules, since tier contents can change and should be confirmed on the Langfuse pricing pages before you commit.
| Factor | Langfuse Cloud | Self-hosted Langfuse |
|---|---|---|
| License and edition | Hosted service on the Langfuse codebase, billed by plan (Hobby, Core, Pro, Teams add-on, Enterprise) | Core features under MIT without limits; ee directories under a commercial license |
| Where traces are stored | In Langfuse Cloud, in the US, EU or JP region you select | In your own PostgreSQL, ClickHouse, Redis or Valkey and S3-compatible storage |
| Available regions | US, EU or JP data regions, plus AWS PrivateLink per the pricing page | Wherever you deploy it, including your own cloud accounts or data centers |
| SOC 2 and ISO 27001 reports | Included on Pro; confirm the current report access process with Langfuse | Not provided by Langfuse for your deployment; your own controls and evidence apply |
| HIPAA-ready region | Available on Pro; confirm the BAA and scope with Langfuse before sending PHI | Depends on how you configure and operate the stack |
| SSO and SCIM | Enterprise SSO and SSO enforcement with the Teams add-on; SCIM API on Enterprise | SSO and SCIM (org management API) require an Enterprise license key |
| Data access window | Hobby 30 days, Core 90 days, Pro 3 years, per the pricing page | Set by your own retention policy; data retention policies need an Enterprise license key |
| Cost driver | Units per billing cycle (traces, observations and scores), plus plan fee and extra units at 8 US dollars per 100k units | No Langfuse usage fee for the open source edition; you pay for infrastructure, staff time and any Enterprise license |
| Who runs the platform | Langfuse operates the service and its uptime SLA is an Enterprise feature | Your team or DeployOpen operates web, worker and data stores, upgrades and backups |
What data location and compliance mean for each model
Langfuse Cloud and self-hosted Langfuse answer the same regulatory questions in different ways. With Cloud, you rely on the provider's regions, reports and plan tiers. With self-hosting, you own the data location and the controls, but you also own the evidence an auditor will ask for. The cards below take each question in turn.
Data location and residency
Langfuse Cloud offers US, EU or JP data regions, so the region you pick at signup determines where traces sit. Self-hosting places every component (web, worker, PostgreSQL, ClickHouse, Redis or Valkey, and blob storage) in the cloud account, data center or network you choose. Check which region your legal team accepts, whether any LLM provider receives trace content outside it, and where backups and blob exports are stored.
Who can access trace data
On Cloud, access to trace data is governed by the provider's staff controls and your organization's roles inside the platform. Self-hosted, access is governed by who holds infrastructure credentials for the database, the object store and the Langfuse instance itself. Map both paths before go-live: list who can read ClickHouse tables, who can read the blob bucket, and who can export projects.
Compliance reports and what they prove
On the Pro plan, Langfuse Cloud lists SOC 2 and ISO 27001 reports. A report is a statement about the provider's controls, issued by an independent auditor, and it does not transfer to your self-hosted deployment. Self-hosting means your own controls and evidence must be assessed by your auditor. DeployOpen prepares systems, controls and evidence for assessments but does not issue audit reports, attestations or certifications.
SSO and RBAC by tier
On Cloud, enterprise SSO, SSO enforcement and fine-grained RBAC sit in the Teams add-on on top of your plan. Self-hosted, the MIT edition covers core features without limits, while project-level RBAC roles and SCIM provisioning need an Enterprise license key. Confirm which identity controls your access policy requires, then price the tier that provides them, and check the authentication and SSO documentation for the exact setup.
AWS PrivateLink on Cloud
Langfuse Cloud offers AWS PrivateLink, which lets your AWS workloads reach the service over private network paths rather than the public internet. This matters when your LLM applications run in a VPC and security policy forbids outbound public traffic for trace data. If you are not on AWS or do not need private connectivity, this control is not a factor. Confirm the PrivateLink setup and the plan it requires with Langfuse before you design the network path.
Retention control
Cloud retention follows the data access window of your plan, while data retention management is listed on Pro. Self-hosted, retention policies are an Enterprise license feature, so without the key you must set your own deletion jobs or accept indefinite storage. Decide your retention period from your policy, then confirm how each option deletes data from ClickHouse and blob storage, including backups.
How Langfuse Cloud unit pricing and data windows add up
Langfuse Cloud bills by unit, and a unit is counted per trace, per observation and per score. The billable units docs define units per billing cycle as the count of traces plus observations plus scores, so a single request that produces one trace, several nested observations and a few evaluation scores consumes well more than one unit. Before you compare plans, pull a week of real traffic from a pilot, count how many observations each request generates on average, and multiply out to a monthly figure. Instrumentation choices, such as how many spans your SDK or OpenTelemetry setup emits, move this number more than any plan label does.
The published plans step up on units and on how far back you can see data. Hobby is free with 50k units per month, 30 days of data access and 2 users. Core is 29 US dollars per month with 100k units and 90 days of data access. Pro is 199 US dollars per month with 3 years of data access, and extra usage is billed at 8 US dollars per 100k units, according to the Langfuse pricing page. Pro is also where data retention management, SOC 2 and ISO 27001 reports and the HIPAA-ready region appear. The Teams add-on at 300 US dollars per month adds enterprise SSO, SSO enforcement and fine-grained RBAC, and Enterprise at 2,499 US dollars per month adds audit logs, the SCIM API, an uptime SLA and a support SLA. Confirm current prices on the Langfuse pricing page before you budget, since these figures are as published.
The self-hosted edition has no license fee for core features, but it is not free to run. Your costs are the cloud accounts or data centers for the web, worker, PostgreSQL, ClickHouse, Redis or Valkey and blob storage, plus the staff who patch, back up and upgrade them. Langfuse's scaling guide lists minimum resources for each component, and its v3 to v4 upgrade guide calls for about 3x ClickHouse disk headroom when backfilling history, so storage and staff time recur. When you request a quote from any implementation partner, ask for the infrastructure sizing per environment, the expected unit volume, the retention period you need, the backup and restore approach, and whether any Enterprise license features are in scope. Ask the partner to state which costs are one-time and which recur each month.
What your team runs in each hosting model
This table sets the day-to-day operating tasks of Langfuse Cloud against a self-hosted Langfuse stack, so you can see who does each job. Read the self-hosted column as the work your team inherits, and check the minimum resource and version facts against your own sizing before you commit.
| Operating task | Langfuse Cloud | Self-hosted Langfuse |
|---|---|---|
| Provisioning | Langfuse provisions and runs the service in the US, EU or JP region you select. | Your team, or a partner you engage, provisions web, worker, PostgreSQL, ClickHouse, Redis or Valkey and S3-compatible storage. |
| Initial deployment target | Not applicable. You create an account and project. | Docker Compose on a single VM is documented for testing with no high availability. Kubernetes with Helm or Terraform is the documented production path on AWS, Azure and GCP. |
| Minimum resources per component | Not applicable to your team. | Web, worker and PostgreSQL: 2 CPU and 4 GiB each. Redis or Valkey: 1 CPU and 1.5 GiB. ClickHouse: 2 CPU and 8 GiB, with at least 16 GiB memory for larger deployments (Langfuse scaling guide). |
| Upgrades within a major version | Langfuse applies platform updates. | Your team applies updates. Database migrations run automatically at startup, which means you need a tested rollback path. |
| Major version upgrades | Langfuse manages the migration. | Your team follows the dedicated migration guide. The v3 to v4 path requires ClickHouse 25.12 or later (26.4 recommended), PostgreSQL 15 or later and Redis 7.0 or later. |
| Backfill capacity during v3 to v4 | Not applicable to your team. | Plan for about 3x ClickHouse disk headroom when backfilling historic data, according to the Langfuse v3 to v4 guide. Confirm the figure against your data volume. |
| Backups and restore | Confirm with Langfuse which backup and restore commitments your plan includes. | Your team backs up PostgreSQL, ClickHouse and blob storage, and tests restores across all of them. |
| Scaling | Langfuse scales the service. Confirm throughput limits for your plan. | Your team scales worker containers on CPU load and, under heavy ingestion, splits langfuse-web into separate ingestion and UI deployments. |
| High availability and on-call | Langfuse operates the service. Confirm the uptime SLA available on your plan. | Your team owns high availability for each stateful component and on-call response. An uptime SLA is part of Enterprise, which is quoted separately. |
How to move traces between Cloud and your own stack
This sequence takes you from choosing a target and region to a verified cutover, so that traces land in one place at a time and you know what history stays behind. It applies in either direction, whether you are leaving Langfuse Cloud for self-hosting or moving the other way.
Decide the target, region and plan
Your security or platform lead records where traces must live, which Langfuse Cloud region (US, EU or JP) applies if the destination is Cloud, and which compliance reports or SSO features the destination must provide. For Cloud, check the tier against the pricing page, since SOC 2 and ISO 27001 reports, the HIPAA-ready region and Teams SSO sit in specific tiers. For self-hosting, note that the MIT edition includes all core features, while RBAC, retention, audit logs and SCIM need an Enterprise license key. Done when the decision is written down with an owner and a named region or infrastructure location.
Stand up the self-hosted stack at a matching version
Your platform team deploys the web and worker containers with PostgreSQL, ClickHouse, Redis or Valkey and S3-compatible blob storage, using Docker Compose for a test VM (no high availability) or Helm or Terraform for production. Pin the Langfuse version to match the release you run on your current source, and check the v3 to v4 guide's prerequisites if you are crossing that boundary: ClickHouse 25.12 or later, PostgreSQL 15 or later and Redis 7.0 or later. Done when the web UI loads, a health check passes and a test trace appears in a project.
Switch SDK or OpenTelemetry endpoints
Your application developers change the host and keys in the Langfuse SDK configuration or the OpenTelemetry exporter so new traces go to the new base URL, with project keys created on the destination. Run this change in a non-production environment first and keep the old credentials available until the cutover is confirmed. Done when a sample request from each service appears in the destination project with the expected traces, observations and scores.
Confirm export options with Langfuse
Before you plan any migration of existing data, ask Langfuse in writing whether it offers an export or import path between Cloud and self-hosted deployments, what formats it supports and what it does not carry over, such as prompts, datasets, scores or user accounts. The sources reviewed do not describe a cross-deployment migration tool, so do not assume one exists. Done when the answer is in writing and you know which artifacts you must recreate or re-export yourself.
Plan historical data and retention windows
Your team lists how much history must move and how much can be left behind. Cloud data access depends on plan: 30 days on Hobby, 90 days on Core and 3 years on Pro, so older traces on Cloud may already be out of reach. Self-hosted retention is governed by your own policies, and retention policy management is an Enterprise feature. If you backfill historic data into a self-hosted ClickHouse, the v3 to v4 guide advises about 3x ClickHouse disk headroom. Done when the retained history, the owner of each dataset and the disk headroom are documented.
Run both stacks in parallel
Your team sends traces to both the old and new endpoints for a defined verification period agreed internally, and compares trace counts, scores and user access in each. Unit-based billing on Cloud means dual writing also adds billable units while both run, so check your unit count against the plan's included units. Done when counts and sample traces match and the on-call team has confirmed that alerts and dashboards read from the new stack.
Plan the cutover and rollback
Your platform lead sets a cutover window, removes the old endpoint from production configuration, and keeps a documented rollback path that restores the previous keys and base URL. Revoke the old credentials only after the new stack has run cleanly for the agreed period, and record who signed off. Done when production traces flow only to the destination, the rollback steps have been tested in non-production, and the change record is filed.
What self-hosting asks of your platform team
Self-hosting Langfuse moves the operating load for trace storage onto your platform team. A self-hosted deployment runs two application containers, Langfuse Web (UI and APIs) and Langfuse Worker (asynchronous event processing), backed by PostgreSQL, ClickHouse, Redis or Valkey, and S3-compatible blob storage. Someone on your team needs working knowledge of each of those stores: backups, disk growth, memory limits and connection handling. 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. Compose on a single VM is documented for testing without high availability, so production planning starts with Kubernetes and an on-call rota that covers all of these components.
The MIT core is complete for tracing, and the Enterprise license key adds governance features. Langfuse states that all core features and APIs are available in the MIT-licensed open source version without limits. The license key unlocks project-level RBAC roles, protected prompt labels, data retention policies, audit logs, server-side data masking, UI customization, organization creators, and the org management API with SCIM. Your compliance lead should map each control your auditors expect to one of these features before deciding whether the key is required. Confirm with Langfuse how the key is issued and validated, and ask what the instance does if the key lapses or is removed, since the answer determines your contingency plan.
Running the open source edition means you carry the paid-feature gap yourself. Cloud features such as SSO enforcement and fine-grained RBAC sit in the Teams add-on, while the self-hosted Enterprise plan adds a named support engineer, solutions architect access and a support SLA, priced by quote. Each of those governance and support items either comes from the license key or from your own engineering. Before you commit, list the audit, retention and access requirements in writing, and ask Langfuse which ones the key covers. Plan for upgrades too: minor updates apply migrations at startup, but major upgrades follow dedicated guides, and the v3 to v4 guide requires ClickHouse 25.12 or later, PostgreSQL 15 or later and Redis 7.0 or later.
Questions teams ask before choosing a hosting model
What does self-hosting Langfuse cost compared with Langfuse Cloud?
Self-hosting removes the per-unit subscription, but you pay for the infrastructure and the people who run it. The open source edition has no license fee for core features, according to Langfuse's self-hosted pricing page. Your recurring costs are the compute and storage for the web and worker containers, PostgreSQL, ClickHouse, Redis or Valkey, and S3-compatible blob storage, plus the engineering time for upgrades, backups and monitoring. Cloud costs rise with trace, observation and score volume, because Langfuse bills units defined as traces plus observations plus scores, with the Hobby, Core and Pro plans priced as listed on its pricing page. Ask any quote to separate one-time setup from recurring operations and to show which volume assumptions it uses.
How long does a self-hosted Langfuse deployment or migration take?
No supplied figures exist for deployment or migration duration, so confirm this before you plan. The duration depends on your target platform, whether you run Docker Compose or Kubernetes, how many trace records you must backfill, and how much of the stack you already operate. Langfuse's v3 to v4 guide advises about 3x ClickHouse disk headroom when backfilling historic data, which is one factor to size early. Ask for a written scope with deliverables and dates before work starts, and ask which tasks sit inside that scope.
Can our own engineers run self-hosted Langfuse without outside help?
Yes, if your team already runs PostgreSQL, ClickHouse, Redis and object storage in production. Langfuse documents Docker Compose on a single VM for testing with no high availability, and Kubernetes with Helm or Terraform on AWS, Azure and GCP for production. Minor updates within a major version apply database migrations automatically at startup, but major upgrades follow dedicated migration guides. Teams without those skills often bring in an implementation partner for setup, hardening and upgrades, or buy ongoing support. DeployOpen works this way on a fixed scope, and the written scope should state which tasks remain with your staff.
Which SSO features and compliance reports sit in which Langfuse Cloud tier?
On Langfuse Cloud, SOC 2 and ISO 27001 reports and data retention management come with Pro. Enterprise SSO, SSO enforcement and fine-grained RBAC come with the Teams add-on, priced at 300 US dollars per month on the pricing page. Audit logs, the SCIM API, an uptime SLA and a support SLA come with Enterprise, priced at 2,499 US dollars per month. Confirm the current feature matrix on the pricing page before you sign, because tiers change. For self-hosting, SSO is documented in the authentication and SSO guide, while RBAC roles, audit logs, data retention and SCIM need an Enterprise license key.
Is EU or JP data residency available on Langfuse Cloud?
Yes. Langfuse Cloud offers US, EU or JP data regions, according to the feature comparison on its pricing page. Pro also lists a HIPAA-ready region. Choose the region when you create the project, and confirm with Langfuse which region your organization currently uses and whether any data leaves it, including support access and sub-processors. If your residency rules require data to stay inside your own accounts or data centers, Cloud does not meet that need and self-hosting is the option to evaluate.
What happens to our trace data when we move between Cloud and self-hosted Langfuse?
Langfuse states that self-hosted deployments use the same codebase as Langfuse Cloud, but it does not promise an automatic export of existing trace data between the two. Plan the move as a data project. Check the data access window on your current plan, since Hobby keeps 30 days, Core 90 days and Pro three years, and decide which history must be carried over. Ask Langfuse and your implementation partner how exports are performed, how long they take at your volume, and how you will verify record counts. Keep the old environment readable until the new one is validated.
When is a Docker Compose setup not enough for production Langfuse?
Docker Compose on a single VM is documented for testing and has no high availability, so it does not meet most production requirements on its own. Move to Kubernetes with Helm or Terraform when you need redundancy, scaling and managed backing services. Langfuse recommends scaling worker containers on CPU load and, under heavy ingestion, splitting langfuse-web into separate ingestion and user interface deployments. The scaling guide lists minimum resources for web, worker, PostgreSQL, Redis or Valkey and ClickHouse, with at least 16 GiB of ClickHouse memory for larger deployments. Check these against your expected volume before choosing the platform.
Questions to answer before you choose
Start with the data in your traces. Prompts, completions and tool outputs often contain customer or patient information, so the first question is whether that data may leave your boundary at all. If it may not, self-hosting is the answer and the rest of this decision is about cost and effort. If it may, check region and report requirements. Langfuse Cloud offers US, EU or JP data regions, and the Pro plan adds SOC 2 and ISO 27001 reports and a HIPAA-ready region, as listed on the Langfuse pricing page. Ask your compliance lead which reports an auditor will request and whether a Cloud plan provides them. Self-hosting moves that evidence work to you. DeployOpen can prepare controls and evidence for an assessment, but audit reports come from independent auditors.
Next, measure your volume in units. Cloud bills every trace, observation and score as a billable unit, so estimate units per billing cycle from a sample of your real pipeline before comparing the Hobby, Core and Pro plans. Extra usage is priced per 100k units. Data access windows also differ by plan, from 30 days on Hobby to 90 days on Core and 3 years on Pro, so check whether the window fits your audit and investigation needs. Self-hosting carries no per-unit fee, but you pay for the infrastructure that absorbs the same volume. Ask for an infrastructure estimate built on your unit count, not on demo traffic. If you need retention controls on your own stack, confirm early that the Enterprise license key covers the data retention policies you require.
How DeployOpen scopes and runs a self-hosted Langfuse deployment
DeployOpen is a Chainscore Labs company that plans, implements, hardens, backs up, upgrades, documents and supports self-hosted open-source platforms inside your own infrastructure. For Langfuse, we start with the written scope that decides whether traces stay in Langfuse Cloud or move to your own stack, then build and hand over the deployment in your cloud account, data center or air-gapped network.
How we scope
- We confirm which Langfuse Cloud region, plan or self-hosted edition your traces and compliance evidence actually need.
- We record the ClickHouse, PostgreSQL, Redis or Valkey and S3-compatible storage you will run, or the managed services you choose.
- We state in writing whether any Enterprise license key features are in scope, since they are not included in the MIT edition.
How we build
- We deploy Langfuse Web and Langfuse Worker in your environment, using Docker Compose for testing or Helm and Terraform for production.
- We configure authentication and SSO, backups and data retention according to the agreed scope and your own policies.
- We test ingestion, retention and access controls against the agreed acceptance criteria before the system goes live.
How we hand over
- We document the deployment, upgrade path and the configuration values your team needs to operate Langfuse without us.
- We prepare the controls and evidence your compliance assessors will ask for, without issuing any audit report or certification.
- We walk your operators through scaling decisions, such as adding worker containers or splitting web deployments under heavy ingestion.
How we support
- Fixed-scope engagements have deliverables and timelines agreed in writing before work starts, so changes go through a written change request.
- Ongoing support, upgrades and continuous testing are available as a retainer that you can scope to your own release schedule.
- Before any major Langfuse version upgrade, we review the migration guide against your ClickHouse, PostgreSQL and Redis versions with you.
