Which LangSmith vs Opik choice fits your team
LangSmith fits teams already on LangChain's Enterprise plan that can accept either a managed cloud region or a licensed self-hosted deployment. Self-hosted LangSmith is an add-on to that plan, requires a license key, and deploys to your Kubernetes cluster with Helm charts, with Terraform also documented. Opik fits teams that need the whole platform, including the server backend, web application, tracing, evaluations and prompt management, running in their own infrastructure under the Apache-2.0 license with no license key. If your renewal is forcing a choice between a managed region and an Enterprise add-on, the question to settle first is whether a license key is acceptable in your environment at all.
The decision turns on four points. The first is data location: LangSmith Cloud keeps data in the region where the account signed up, while Opik can sit wholly in your own accounts or data centers. The second is the license needed for self-hosting, which determines whether procurement must sign an Enterprise contract before any deployment work starts. The third is retention ownership: LangSmith retention follows your plan, while Opik retention is set by the storage you run, so your ClickHouse sizing and cleanup policy become your own responsibility. The fourth is in-house operating capacity. Opik runs on ClickHouse, MySQL and Redis, so someone must own backups, upgrades and capacity planning for those components. Ask your LangChain account team for the Enterprise terms in writing, and ask your platform team to name who would run the Opik stack day to day before you commit to either path.
Which option fits which scenario
This table maps common LangSmith and Opik situations to the option that usually fits better and why. Read each row as a condition to test against your own contracts, network rules and data requirements, not as a verdict. The main takeaway is that Opik's self-hosting without a license key matters most when the constraint is licensing or egress, while LangSmith Cloud regions and plan-based retention still matter for teams already committed to LangChain.
| Scenario | Better fit | Reason and what to confirm |
|---|---|---|
| Air-gapped network with no outbound access | Opik | Self-hosted Opik runs on infrastructure you control and needs no license key. Confirm which container images and model endpoints you must mirror before go-live. |
| Existing LangChain Enterprise contract that already includes self-hosting | LangSmith | Staying on LangSmith avoids a re-instrumentation project if your license covers self-hosting. Confirm with LangChain whether the license key and Kubernetes deployment are in scope. |
| Stack built on LangChain or LangGraph without an Enterprise plan | Opik | Opik has native LangChain and LangGraph integrations and self-hosts without an enterprise add-on. Confirm callback changes in a test project first. |
| Non-LangChain stack using OpenTelemetry or several frameworks | Opik | Opik supports OpenTelemetry and has integrations with many frameworks plus a decorator for custom code. Check that your framework versions are covered by the current integration list. |
| Retention beyond the plan's trace limits | Opik | Self-hosted retention is set by your own ClickHouse storage and policy rather than by a plan tier. Confirm the storage growth and backup plan for the retention period you need. |
| EU data residency requirement | Opik | A self-hosted deployment in an EU data center keeps traces and prompts in region, while LangSmith Cloud's EU region still keeps billing information in the US. Confirm with your data protection team whether billing data is in scope. |
| Need for LangSmith's hosted operations with no infrastructure team | LangSmith | LangSmith Cloud removes database and upgrade work from your team. Opik's hosted option on Comet.com is the closer comparison, so ask for its region and retention terms. |
| Need for enterprise SSO, RBAC and audit features | Depends on contract | Verify which features each plan and self-hosted edition includes in writing before you decide. Ask both vendors for the current feature matrix. |
Where LangSmith and Opik keep your trace data
LangSmith Cloud runs in four regions: GCP US (Iowa), GCP EU (Netherlands), GCP APAC (Sydney) and AWS US (Ohio). Trace data stays in the region where the account signed up, but billing information stays in the US, so a single account does not keep every category of data in one jurisdiction. Your team does not operate the infrastructure, which shifts patching, scaling and backup of the data plane to LangChain.
- Region choice: Confirm the region at sign-up, since it determines where trace data is stored.
- Billing data: Billing information is kept in the US even for EU accounts, so include it in your data mapping.
- Identity: Map SSO and role assignments to your identity provider, and confirm which plan includes them.
- Network egress: Your application sends traces outward to the vendor endpoint, so review outbound rules for each environment.
- Evidence: Request the vendor's documentation and contract terms on data handling. DeployOpen does not issue audit reports; independent auditors do.
Before you approve this option, confirm the regional storage terms, the billing data location and the subprocessors in writing with LangChain.
Self-hosted LangSmith is an add-on to LangChain's Enterprise plan. It needs a license key and deploys to your Kubernetes cluster with Helm charts, with Terraform also documented. Trace data then sits in your cluster and storage, which keeps it in your control, but the license is an ongoing commercial dependency tied to an Enterprise contract.
- Data location: Traces, datasets and prompts live in the storage your cluster uses, under your own region and retention choices.
- License key: Confirm with LangChain how the key is validated and what happens to running deployments if the license lapses.
- Identity: Plan SSO, group mapping and service accounts inside your cluster, and confirm which of these depend on the Enterprise tier.
- Network egress: Ask LangChain which endpoints the deployment contacts for licensing and telemetry, and whether they can be blocked or proxied in your environment.
- Operations: Your platform team owns the Kubernetes lifecycle, backups and upgrades of the self-hosted stack.
Ask for the exact list of outbound endpoints and the license validation behavior in writing before you design the network policy.
Opik is Apache-2.0 licensed, and its README states the full platform, including tracing, evaluations and prompt management, can be self-hosted without a commercial license. You can run it locally with Docker Compose through its opik.sh script, or on Kubernetes with Helm charts for production. Its infrastructure uses ClickHouse, MySQL and Redis, so each of those stores becomes part of your data review.
- Data location: Traces, evaluation results and prompts are stored in your ClickHouse, MySQL and Redis instances, inside your own network.
- Identity: Authentication and access control depend on how you configure the deployment, so confirm the supported options for your environment.
- Network egress: Self-hosting does not require a license check call, so you decide which outbound paths, if any, to allow, including for model providers.
- Compliance evidence: Document backup, encryption at rest and access logs for ClickHouse, MySQL and Redis, since each holds trace content.
- Scope: DeployOpen can prepare configuration, hardening and evidence for these controls; independent auditors issue any formal report.
Confirm the authentication options, encryption settings and log retention for each datastore against your own control requirements before go-live.
What the Enterprise add-on license changes for self-hosting
Self-hosting LangSmith is not a free option in the way self-hosting Opik is. According to LangSmith's self-hosting documentation, self-hosted LangSmith is an add-on to LangChain's Enterprise plan, it requires a license key, and it deploys to your own Kubernetes cluster with Helm charts, with Terraform also documented. The license key is the gate: the software runs in your infrastructure, but the right to run it depends on an enterprise contract with LangChain. Opik, by contrast, is Apache-2.0 licensed, and 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. Before you compare the two on features, decide whether a commercial contract is a cost you want to carry for this component.
Put the following questions to LangChain in writing before you decide, and do not rely on verbal assurances. First, confirm what the license key permits: number of clusters, environments (production, staging, disaster recovery), and whether a license covers air-gapped or offline operation. Second, confirm the support scope: which components are covered (the application, the bundled data stores, Helm chart changes), response channels, and whether support includes help with upgrades and incident triage in your own cluster. Third, confirm upgrade rights: whether the license includes access to new chart versions and patch releases for the term, and what happens to a running deployment if the contract lapses or is not renewed. Ask also what telemetry or license validation the deployment performs. Compare the answers with Opik's Apache-2.0 terms, which your legal team can review directly in the repository's license file, and with the release cadence you can verify on Opik's releases page.
What drives the cost of each option
The two options bill different things. LangSmith's self-hosted path is tied to an Enterprise contract, while Opik's cost sits in the infrastructure you run and the people who operate it. The cards below separate recurring from one-time costs and list what to request before comparing quotes.
Enterprise add-on and license key
Self-hosted LangSmith is an add-on to LangChain's Enterprise plan and needs a license key, so the commercial subscription is a recurring line that sits alongside your infrastructure. Ask whether the quote covers self-hosted seats, the license key renewal term, support levels and any usage-based charges. For Opik, the Apache-2.0 license carries no license fee, so the comparable line is the cost of running the platform yourself, plus an optional hosted subscription on Comet.com if you choose that route.
Kubernetes and cloud infrastructure
Both self-hosted deployments run on Kubernetes, LangSmith through Helm or Terraform and Opik through Helm charts for production. Compute, load balancing, ingress, and cluster administration are recurring costs in your cloud account or data center, and they scale with trace volume and the number of concurrent users. Confirm which node sizes and replica counts the vendor documentation assumes for your expected trace rate before you size the cluster.
ClickHouse and storage growth
Opik stores trace data in ClickHouse, so storage cost grows with trace volume and with how long you keep data. Retention on your own cluster is a decision you make, not a plan setting, which means storage, backup and compaction costs recur for as long as traces are kept. Ask your team to model retention periods against expected daily trace volume, and to confirm disk, backup and replication requirements for the ClickHouse cluster.
MySQL and Redis operations
Opik also depends on MySQL and Redis, two additional stateful services with their own backup, patching and failover needs. These are mostly one-time setup costs followed by recurring operations time, and they add to the on-call surface your platform team owns. In a quote, ask who builds the backup and restore runbooks for each store and whether managed database services from your cloud provider are an accepted substitute for in-cluster deployments.
Upgrades and version management
Self-hosting means you schedule and test every upgrade, including schema migrations across ClickHouse, MySQL and the application services. On the LangSmith side, upgrades follow the Enterprise support process, and on the Opik side, releases are published on GitHub and tracked by your own team. Recurring upgrade effort depends on how often you adopt new releases and how much of your tracing and evaluation logic is customized, so ask for the expected upgrade process and staging support in any quote.
Internal or partner support time
Support cost is where the two options differ most in practice. A LangSmith Enterprise contract brings vendor support as part of the subscription, while Opik's support comes from your own engineers, the community, or a partner under a fixed-scope or retainer agreement. Estimate the internal hours for on-call, incident response and integration work, and ask any partner to state in writing which tasks, response expectations and deliverables the retainer covers.
How a move from LangSmith to Opik usually runs
The sequence moves tracing, prompts and evaluations to a self-hosted Opik platform in stages, so your team can compare results against LangSmith before any traffic or history is cut over.
Inventory LangChain and LangGraph callbacks
Engineers list every application, service and notebook that sends traces to LangSmith, and note how each one is wired: LangChain tracing environment variables, LangGraph runs, custom code that calls the LangSmith client directly, and any OpenTelemetry exporters. The owner of each service signs off on the list. The step is done when every trace source is named, assigned an owner and marked as either a native integration or custom instrumentation that needs rework. Effort depends mostly on how many services and how much custom code call the LangSmith SDK directly.
Stand up Opik with Docker Compose for evaluation
The platform team starts Opik on a non-production host using the Docker Compose setup that the Opik README describes, launched through its opik.sh script. The stack includes ClickHouse, MySQL and Redis, so the team checks that the host has the CPU, memory and disk the project documentation specifies for the expected trace volume. A developer sends a test trace from a LangChain or LangGraph app and confirms it appears in the web application. The step is done when a known test run is visible end to end and the team has noted which configuration values they changed.
Deploy the production Helm release
Once the evaluation instance behaves as expected, the platform team moves the same configuration to Kubernetes using the Helm charts Opik publishes for production. They decide on the storage class for ClickHouse, backup approach for MySQL and ClickHouse, ingress, TLS and how secrets are injected from the cluster's secret store. Security reviews the network paths so that no component needs egress for its own operation. The step is done when the release is healthy, backups have been restored once in a test, and access is limited to the agreed identity provider or user list.
Switch integrations and instrument custom code
Developers change the tracing configuration in each service from the inventory, pointing LangChain and LangGraph applications at the Opik instance and using the native integrations. Custom code that relies on a decorator or manual spans is moved to Opik's @track decorator or to OpenTelemetry export. Each change is made in a feature branch and deployed to a staging environment first. The step is done when every service in the inventory produces traces in the self-hosted project, with span hierarchy matching the earlier LangSmith view for a sample of requests.
Rebuild evaluators, datasets and prompts
The evaluation owners recreate the LLM-as-a-judge metrics, datasets and online evaluation rules they used in LangSmith, using Opik's evaluation metrics and prompt management with its playground. They check each rebuilt metric against a set of runs whose outcomes are already known, so a change in judge behavior is not mistaken for a model regression. Prompt versions are loaded and linked to the traces that use them. The step is done when the team signs off that each metric reproduces its earlier verdicts on the reference runs, and any metric that does not is documented as a gap.
Export LangSmith data while plan access lasts
The data owner requests a LangSmith bulk export, which writes trace data to an S3-compatible bucket in Parquet format. Confirm with LangChain whether your account keeps access to bulk export under the current plan terms, since the researched facts say access depends on when the account signed up and the plan tier. The team then maps Parquet fields onto Opik projects and imports the historical records that matter for audits and regression baselines. The step is done when a sample of exported traces has been loaded into Opik and compared field by field with the originals.
Run both platforms in parallel and retire LangSmith
For a defined period, services send traces to both LangSmith and Opik, and the team compares dashboards, evaluation scores and cost or token counts for the same requests. Discrepancies are logged and resolved before any service is removed from the LangSmith configuration. Retirement covers revoking API keys, removing the tracing environment variables from deployment manifests, and confirming with the vendor how long data remains in the account. The step is done when the parallel comparison has no open discrepancies and the owner has written down the LangSmith data retention and deletion status.
Questions buyers ask before choosing Opik or LangSmith
Can Opik replace LangSmith evaluations and prompt management?
Opik covers both areas in its open-source release. Its README lists LLM-as-a-judge metrics for hallucination, moderation and RAG, online evaluation rules, and prompt management with a playground, and states that the full platform can be self-hosted without a commercial license. Confirm the fit by listing the LangSmith features your team uses today, such as datasets, experiments, annotation queues and the prompt hub, and mapping each one to an Opik equivalent in a test project. Check whether your current evaluators depend on LangSmith-specific APIs, since those must be rebuilt rather than copied.
What does self-hosting LangSmith require?
Self-hosted LangSmith is an add-on to LangChain's Enterprise plan and needs a license key. It deploys to your Kubernetes cluster with Helm charts, and Terraform is also documented. Before you start, ask LangChain to confirm the license terms, the cluster sizing it expects, and whether the deployment reaches its license service during operation, since that affects air-gapped plans. The recurring costs are the Enterprise contract, the cluster and the people who operate it.
What cost drivers apply when moving from LangSmith to Opik?
The license cost of Opik is zero because it is Apache-2.0, so your spend moves to infrastructure and operations. Opik self-hosts on ClickHouse, MySQL and Redis, so storage and compute grow with trace volume and retention. The main cost lines are:
- Compute and storage for the ClickHouse, MySQL and Redis services, scaled to trace volume.
- Retention, because longer retention means more stored data.
- Staff time for upgrades, backups and on-call response.
- Migration work, including re-instrumenting code and rebuilding evaluators.
Ask any vendor quote to split these lines so you can compare them with your current LangSmith invoice.
How long does a LangSmith to Opik migration take?
No duration is fixed in the supplied facts, so the timeline depends on your scope. It is driven by the number of LangChain and LangGraph applications, the count of evaluators and datasets to rebuild, the prompts to move, and how much historical trace data you keep. Opik supports native LangChain and LangGraph integrations, which shortens instrumentation work compared with custom code. Ask for a written plan that sequences a pilot project, parallel running and cutover, and confirm the effort with a scoped assessment of your environment.
What does the migration need from our engineering team?
Your team needs people who know your LangChain or LangGraph code, someone who can run Kubernetes or Docker Compose, and an owner for the ClickHouse, MySQL and Redis services. Opik offers Docker Compose through its opik.sh script for local use and Helm charts for production, so the platform skill set is standard. Plan time for changing callback setup in each application, rewriting evaluators that use LangSmith APIs, and validating that dashboards and alerts still work. Confirm these responsibilities in writing before you start.
Can our team run the migration alone?
Many teams can, especially when they have a Kubernetes platform team and a few engineers familiar with the applications. The harder parts are rebuilding evaluation logic, agreeing on retention and access controls, and operating the stack after go-live. If those skills are thin, a fixed-scope engagement can cover the platform setup, integrations, hardening, backups and documentation, with ongoing support available as a retainer. DeployOpen sets scope, deliverables and timelines in writing before work starts, and does not publish a price list.
How are historical LangSmith traces handled during the move?
Historical traces are not moved automatically by the Opik integration, so decide early what you need to keep. LangSmith bulk export writes trace data to an S3-compatible bucket in Parquet format, but for customers who signed up after August 3, 2026, that export is only on the Enterprise plan, while earlier Plus customers keep access until February 1, 2027. Confirm your plan and export window with LangChain, export what you need in time, and decide whether older data is archived for audit or imported into the new project. Also confirm your retention needs for compliance reviews.
When staying on LangSmith is the right call
LangSmith remains a sound choice when your application is already built on LangChain or LangGraph, your data can live in one of LangSmith's cloud regions, and your organization values a vendor it can call for support. In that situation, the cost of moving tracing, datasets and prompt management is usually higher than the licensing gain, and the integration is the least friction you will get. Staying is a deliberate decision, so test the conditions rather than assume them.
Start with the region check. LangSmith Cloud runs in GCP US (Iowa), GCP EU (Netherlands), GCP APAC (Sydney) and AWS US (Ohio), and data stays in the region where the account signed up, while billing information stays in the US. Confirm with your privacy and legal team that billing metadata is acceptable and that your region is one of these. Then check the plan terms: the bulk export of trace data as Parquet to an S3-compatible bucket is limited to the Enterprise plan for accounts that signed up after August 3, 2026, so if you depend on exporting history, confirm your plan and your exit path in writing before you commit.
Test the stay decision with a short, time-boxed exercise. Ask LangChain for the current retention terms for your plan, the seat and trace pricing tied to your volume, and the support tier included in your contract, then compare those answers with what you would have to build and run yourself. If the answers hold and your evaluators already live in LangSmith, document the decision, set a review date, and keep a current export of traces and datasets so that a future move remains possible. Revisit the choice if your data residency needs change or if you need self-hosting without an enterprise contract.
