European Alternatives to LangSmith
Looking for a European alternative to LangSmith? LangSmith is sold by LangChain, Inc., a Delaware corporation, under Delaware law, and its regions FAQ states that LangChain has no legal entity in the EU for customer contracting. Pricing is per seat ($39 per seat per month on Plus) plus pay-as-you-go traces, and self-hosted or hybrid deployment is listed under the Enterprise plan.
The European tools below cover tracing, evaluation and agent testing with contracting parties in the Netherlands, Spain, Germany, France, Israel and Sweden, and several of them are open source or can run inside your own network.
How we rank these tools — 4-step process
-
1
European ownership, verified
The company is headquartered and incorporated in the EU, EEA, the UK, Switzerland or Israel, and processes customer data in Europe or Israel. A US parent company disqualifies a tool from this page regardless of where its servers are.
-
2
Category fit and hands-on review
What the tool actually does, who it suits, and where it falls short — checked against the vendor’s own documentation, changelog and pricing page rather than its marketing copy.
-
3
Compliance and pricing check
GDPR posture, hosting location and the prices quoted on this page are verified against the vendor’s public pricing before publication, and re-checked when we revisit the category.
-
4
Position on this page
Placement on this page can be paid, and that can affect which tools appear here and the order they appear in. It never buys a good review: a tool that fails the checks above is not here at any price, and payment does not change the shortcomings we write about. A vendor can ask us to correct a factual error — not to remove a criticism.
Vendors can pay for visibility on this page. It never changes what an entry says about a product, including the criticism, and we earn nothing when you click through to a vendor. Vendors can pay for more visibility, higher up on the review pages; it never affects which tools are listed. Editorial policy
10 European Alternatives to LangSmith
LangWatch
Open-source agent tracing, evaluations and simulations from Amsterdam
Latitude
Open-source agent observability that groups failures into issues
Galtea
Test generation, simulation and evaluation for regulated industries
Giskard
Red teaming and evaluation for LLM agents before release
Orq.ai
AI gateway with built-in traces, evaluation and governance
Rhesis AI
Open-source test sets and expert feedback for AI agents
Coralogix AI Center
AI evaluators and session tracing inside a full observability stack
Qualifire
Guardrails, evaluations and observability using small judge models
Opper
EU-hosted AI gateway with logs, traces and per-call checks
ZenML
Pipeline orchestration plus Kitaru replay of agents on production data
Key takeaways
- LangChain, Inc. is a Delaware corporation, and the LangSmith regions FAQ says it has no EU legal entity for customer contracting.
- LangSmith Plus costs $39 per seat per month with 10k base traces, and every plan is billed in US dollars.
- LangWatch (Netherlands), Latitude (Spain) and Rhesis AI (Germany) publish open-source code, so tracing and testing can run without a hosted contract.
- Galtea and Giskard start from tests and red teaming, which LangSmith treats as one module among many.
- Orq.ai and Opper are gateways whose traces come from routed traffic, and Coralogix adds LLM evaluators to a general observability platform.
Why people leave LangSmith
LangSmith is the tracing, evaluation and deployment platform sold by LangChain, Inc., the company behind the LangChain and LangGraph frameworks. The terms of service, last updated on 2 June 2026, are an agreement between the customer and LangChain Inc., a Delaware corporation. For many agent teams it is the default because the SDK sits one import away from code they already wrote.
The reason European buyers look past it has little to do with features. A trace stores the prompt, the model answer and every tool call, which means it often holds customer names, contract clauses or support conversations. The LangSmith regions FAQ says that data can be hosted in an EU instance, and also says that LangChain has no legal entity in the EU for customer contracting today.
So the data may sit in Europe while the party you sign with, and the law that governs the relationship, stays American. Procurement and privacy teams treat those as separate questions. The tools on this page answer the second one with a contracting party in the Netherlands, Spain, Germany, France, Sweden or Israel, and several can also be run inside your own network.
- No EU entity to contract with The regions FAQ answers the question "Do you have a legal entity in the EU that we can contract with?" with "We do not have a legal entity in the EU for customer contracting today". It adds that the terms are identical in every cloud region. A European data processing agreement is therefore still signed with a Delaware company, whichever region hosts the traces.
- Seats plus metered traces The Plus plan is $39 per seat per month with 10k base traces included, then pay as you go. Usage on the platform is metered in LangChain Standard Units at $1.00 each, and deployments, sandboxes and tuned evaluators draw on the same unit. Costs rise with team size and with trace volume at once, so a forecast needs both numbers.
- Dollar billing and fixed regions The FAQ states that all LangSmith plans are paid in USD and that pricing is the same in each region. An organisation cannot hold workspaces in two regions, and migration between regions is not supported at the moment. A company that picks the US instance by mistake cannot move later without starting a fresh organisation.
- Self-hosting is an Enterprise feature Self-hosted and hybrid deployment appear on the Enterprise row of the pricing page, with custom pricing. In the hybrid model the terms say traces and run data from the customer workloads are sent to LangChain cloud for observability. Teams that want the data to stay entirely in their own cluster have to read that definition carefully.
What you have to replace, not just match
Replacing LangSmith means replacing more than a trace viewer. The pricing page lists tracing, monitoring, insights, online and offline evaluations, dataset collection, annotation queues, a prompt hub and playground, and bulk data export under one product name, and the same subscription also covers Deployment, Fleet, Engine, Sandboxes and an LLM Gateway. Most teams use two or three of these. Write down which ones your engineers open weekly before you compare tools.
Next come the integrations. LangSmith is closest to LangChain and LangGraph code, where tracing needs little more than an environment variable. A European tool will usually ask for an OpenTelemetry exporter or its own SDK decorator instead. That is a few lines per service, but it touches every service that emits traces, and any custom evaluator that reads LangSmith run objects must be rewritten against the new data model.
Finally there is history. Datasets, annotated examples and prompt versions are the assets that took months to build, and traces older than your retention window are probably gone already. Use the bulk data export that the pricing page lists under LangSmith Observability and Evaluation before you cancel, and decide which datasets are worth importing as regression tests and which can be left behind.
The alternatives compared
| Position | Tool | Headquarters | Pricing | Jurisdiction |
|---|---|---|---|---|
| #1 | LangWatch | Amsterdam, Netherlands | Free Developer plan (50k events/month); Growth EUR 29 per core seat per month; Enterprise on request | EU (Netherlands) |
| #2 | Latitude | Barcelona, Spain | Free Starter (20K credits/month); Pro $99/month (100K credits); Enterprise on request | EU (Spain) |
| #3 | Galtea | Barcelona, Spain | Free plan (EUR 0, 1,500 credits/month); Enterprise on request | EU (Spain) |
| #4 | Giskard | France | Open source free; Professional on request | EU (France) |
| #5 | Orq.ai | Amsterdam, Netherlands | Pay-as-you-go: 4.5% fee on Orq-managed keys; 1M free BYOK requests a month, then 4%; Enterprise on request | EU (Netherlands) |
| #6 | Rhesis AI | Potsdam, Germany | Free cloud allowance and free self-hosted edition; see website for paid cloud plans | EU (Germany) |
| #7 | Coralogix AI Center | Ramat Gan, Israel | AI data $1.50 per 1M tokens; logs $0.42 per GB; traces $0.16 per GB; 14-day free trial | Israel (outside the EU/EEA) |
| #8 | Qualifire | Israel (city not stated), Israel | Free plan; Pro $550/month; Enterprise on request | Israel (outside the EU/EEA) |
| #9 | Opper | Sweden, Sweden | Pay as you go: 3% platform fee (Gateway), 5.5% (Control Plane); Enterprise custom | EU (Sweden) |
| #10 | ZenML | Munich, Germany | Open source free; Scale SaaS $999/month; Enterprise on request | EU (Germany) |
How each alternative compares to LangSmith
Best for: Teams that want LangSmith-style tracing and evaluation, but open source and under Dutch terms
LangWatch is published by Reasoning Engines B.V. in Amsterdam, and the terms follow the Dutch Civil Code. It groups tracing, evaluators, agent simulations and prompt management in one workspace, which is the part of LangSmith most teams use daily. The Apache-2.0 code means a managed account and a self-hosted cluster are the same product.
Against LangSmith, the money is easier to read: the Developer plan is free with 50k events a month, and Growth is EUR 29 per core seat with 200k events included, then EUR 5 per 100k. Simulations and evals are unlimited on the paid plan instead of being metered. Retention is shorter, at 14 days on Developer and 30 days on Growth.
What LangWatch does better than LangSmith
- Contracting party in the EU, with an EU cloud region
- Apache-2.0 code you can run in your own VPC
- Evals and simulations not metered separately
- Euro prices on the pricing page
Where LangWatch is a step down from LangSmith
- Data retention is only 14 to 30 days unless storage is bought
- No deployment hosting, sandboxes or no-code agent builder like the LangSmith suite
- Gateway and Langy features are new
Standout against LangSmith. It is the only tool here that combines Apache-2.0 tracing with user simulation that runs inside pytest.
Best for: Teams with live agents who would rather triage recurring problems than read single traces
Latitude Data S.L. in Barcelona builds around what happens after traces arrive. Failed sessions become signals, similar ones are merged into an issue with a trend and example traces, and alerts go to Slack, email or webhooks. LangSmith offers a comparable idea through its Engine module, which clusters agent behaviour into issues, but that module is metered in units.
Ingestion accepts the vendor SDK or any OpenTelemetry pipeline, and the MIT code sits on GitHub. Starter is free with 20K credits and 30-day retention, and Pro is $99 per month with 100K credits and 90-day retention for unlimited seats. Seat-based billing disappears, but credit metering replaces it.
What Latitude does better than LangSmith
- Unlimited seats on every plan instead of per-seat billing
- MIT licence and European servers for hosted data
- Failure clustering included in the standard plans
- Spanish contracting party under Spanish law
Where Latitude is a step down from LangSmith
- Dollar pricing and credit metering that take a pilot to estimate
- Self-hosting support sits on the Enterprise plan
- No deployment or sandbox products
Standout against LangSmith. Failures are grouped into trended issues by default, rather than left as a list of individual runs.
Galtea
Specification-driven test generation for regulated sectors, instead of trace-first tooling
Best for: Banks, insurers and public bodies that must document how an AI system was tested
Galtea Solutions S.L. comes out of research at the Barcelona Supercomputing Center, according to its about page. You describe what the system must and must not do, and the platform produces test cases, simulated multi-turn conversations and red-team attempts. LangSmith expects you to bring a dataset, so the starting point is different.
Credit costs are published per operation: one credit for a red-team or scenario case, two for a judge evaluation, three for a simulated turn. The free plan gives 1,500 credits a month with 90-day retention. Everything above that, including private-tenant or self-hosted deployment, is Enterprise and priced on request.
What Galtea does better than LangSmith
- Generates test sets from policies when no dataset exists yet
- Audit trail and ISO 27001 claim aimed at regulated buyers
- Per-operation credit costs are published
Where Galtea is a step down from LangSmith
- Only the free plan has a public price
- Not open source, and the free plan is cloud only
- Less suited as a general trace explorer
Standout against LangSmith. Test cases are generated from your written requirements, so an evaluation can start before any production traffic exists.
Best for: Teams shipping LLM agents that need repeatable adversarial tests in CI
GISKARD AI SAS, a Paris company founded in 2021, publishes the Giskard library under Apache-2.0. It tests for prompt injection, data disclosure, hallucination and inappropriate refusals, and the Professional tier lists more than 50 adversarial probes, including multi-turn attacks. This covers the red-team part of LangSmith evaluation, in a dedicated tool.
It does not trace production traffic, so most teams run it beside another observability tool. Deployment on-premise, in a private cloud or as SaaS with data residency is listed for the enterprise tier, and the Professional price is not published. The v3 release is a rewrite, so scripts for earlier versions need migration.
What Giskard does better than LangSmith
- Adversarial probe coverage built for security review
- Apache-2.0 library and French vendor
- On-premise and private-cloud deployment options
Where Giskard is a step down from LangSmith
- No tracing, monitoring or prompt management
- Professional pricing is not published
- Open-source tier gets best-effort support
Standout against LangSmith. It targets the failure modes a security team asks about, such as prompt injection and data disclosure, with tests you can rerun on every release.
Best for: Companies that want one governed endpoint for many models, with traces included
Orq.AI Holding B.V. is registered with the Dutch chamber of commerce and has an Amsterdam address. Every request passing through its OpenAI-compatible endpoint is recorded with cost, tokens and latency, so nothing extra has to be instrumented. LangSmith now sells a gateway as well, so the two overlap more than they used to.
The privacy notice says information is processed in the EEA, and the service terms are governed by Dutch law. Pay-as-you-go includes 100k spans a month, then EUR 7 per 100k, with a 4.5 percent fee on models used through Orq-managed keys. On-prem and VPC deployment belong to Enterprise.
What Orq.ai does better than LangSmith
- Budgets, guardrails and PII redaction in the same layer as traces
- Dutch law and EEA processing
- No code changes to start collecting traces
Where Orq.ai is a step down from LangSmith
- Traces cover only traffic that passes through the gateway
- Not open source
- The privacy page does not name a cloud region
Standout against LangSmith. Policy, spend limits and tracing are applied at one gateway call instead of in three separate tools.
Best for: Teams where lawyers, support leads or clinicians know what a right answer looks like
Rhesis AI GmbH is based in Potsdam according to its imprint and README. Reviewers open a shared link, talk to the live agent, turn conversations into test cases and leave verdicts on single turns. Recurring feedback becomes tests that run on each change, which LangSmith handles through annotation queues.
The documentation also lists conversation simulation, adversarial testing, DeepEval-based metrics and tracing with auto-instrumentation. The repository carries an MIT badge plus an enterprise-licensed part, and paid cloud prices could not be verified from the pages that render without scripts.
What Rhesis AI does better than LangSmith
- Non-engineers can review and label without an account in your codebase
- Free self-hosted edition
- German vendor with SSO and audit-log streaming listed
Where Rhesis AI is a step down from LangSmith
- Paid cloud pricing is unclear
- Not a high-volume production trace store
- Licence mixes MIT with an enterprise part
Standout against LangSmith. Every verdict is attached to a concrete test case and agent version, so feedback becomes a regression test.
Best for: Companies that already want logs, metrics, traces and AI agents in a single product
Coralogix Ltd. is incorporated in Israel with its principal place of business in Ramat Gan, and Israel counts as European on this site. The AI Center assigns evaluators for hallucination, PII leaks and toxicity to each agent and scores messages in real time. Documentation lists EU1 on AWS Ireland and EU2 on AWS Stockholm.
Pricing is by volume: AI data at $1.50 per million tokens, logs at $0.42 per GB and traces at $0.16 per GB, with a 14-day trial. The site footer shows Coralogix Inc., and the terms let an affiliate be the contracting party, so confirm the entity on your order form.
What Coralogix AI Center does better than LangSmith
- Follows an incident from a failed model call to the infrastructure below it
- Unlimited users, SSO and audit trail on every account, according to the pricing page
- EU regions in Ireland and Stockholm
Where Coralogix AI Center is a step down from LangSmith
- More platform than a team needs for LLM tracing alone
- No self-hosted edition on the pages checked
- Contracting entity must be confirmed per order
Standout against LangSmith. It is the only option here that puts LLM evaluator results next to ordinary infrastructure logs and metrics.
Best for: Teams that want to block bad output during the request, not just review it afterwards
Qualifire Ltd. issues terms governed by the laws of the State of Israel, and the privacy policy mentions facilities in Israel and the European Commission adequacy decision. Small judge models check prompt injection, PII, hallucination and grounding on each request, and results feed an observability view and a prompt studio.
Plans run from a free tier to Pro at $550 per month with 1 billion monthly tokens, unlimited protection rules and 90-day log retention. The pricing page contradicts itself on free token limits, 3 million in one place and 300k in another, so confirm the number before relying on it.
What Qualifire does better than LangSmith
- Enforces policy in the request path rather than after the fact
- BYOC or on-premise deployment on Enterprise
- Seven-year log retention on Enterprise
Where Qualifire is a step down from LangSmith
- Dollar pricing and a $550 step to Pro
- Thin as a trace explorer for debugging
- Inconsistent free-tier figures on the pricing page
Standout against LangSmith. Guardrails run as small judge models in the request path instead of as a separate review step.
Best for: Teams that want EU-routed model access first and basic trace debugging with it
Opper Technology AB is registered in Sweden under organisation number 559446-3720. Logs list every provider call with model, status, tokens, cost, latency and region, and a trace is a tree of spans for model calls, tool calls and the rules that fired. A trace can be replayed in the playground.
By default only metadata is stored, and inputs and outputs are kept only if a retention rule says so. Gateway is pay as you go with a 3 percent platform fee, and Control Plane is 5.5 percent, with full traces kept for 1 to 30 days. There are no dataset or experiment tools.
What Opper does better than LangSmith
- Zero data retention by default
- EU-hosted API with EU routes
- Per-call checks that flag, redact, block or score
Where Opper is a step down from LangSmith
- No datasets, experiments or annotation queues
- Proprietary, with no self-hosted edition
- Storing traces needs the Control Plane tier
Standout against LangSmith. The default is to store nothing but metadata, so retaining prompts is a deliberate choice.
Best for: ML engineers who care more about repeatable pipelines than about trace dashboards
ZenML GmbH is registered in Munich and governed by German law. It orchestrates Python pipelines across the orchestrators and clouds you already run, and stores artifacts and metadata for lineage. The Apache 2.0 core can be self-hosted.
It is the least direct substitute on this page. ZenML sits beside Kitaru, a separate product for recording and replaying agent sessions, but it has no trace explorer or evaluator suite of the LangSmith kind. Scale SaaS costs $999 a month, so most teams use the free open source plan.
What ZenML does better than LangSmith
- Pipelines move between orchestrators without rewrites
- SOC 2 Type II and ISO 27001 listed on the site
- Free open source plan with no execution cap
Where ZenML is a step down from LangSmith
- No LLM trace explorer or evaluator library
- Jump to $999 per month on Scale
Standout against LangSmith. Every pipeline run stores its artifacts, which gives lineage from a model back to the data that produced it.
What actually breaks when you switch
The first cost of leaving LangSmith is instrumentation. Tracing in LangChain code often rests on environment variables, and replacing it means adding an OpenTelemetry exporter or a vendor decorator to each service. Do it behind a flag, send traces to both systems for a week, and compare span trees before turning the old one off. Pay attention to metadata such as user IDs and session IDs, which different tools name differently.
The second cost is the contract and the people. Terms with LangChain end on their own schedule, so check your order form for notice periods before you sign a new one. Engineers who live in the LangSmith playground and prompt hub need a new place for prompts, and the evaluators that read LangSmith runs must be rewritten as judge prompts or metrics. Export datasets in bulk first, because regions cannot be merged and retention may have removed older runs.
Does choosing the EU region of LangSmith solve the contracting question?
It solves where the traces are stored, not who you contract with. The regions FAQ lists a US, an EU and an APAC instance, and confirms that a company outside a region can still host there. It then says the same legal terms apply in every region and that there is no EU entity for customer contracting.
If your policy only requires EU storage and a DPA, the EU instance may be enough, and LangChain says a DPA can be requested through support. If your policy requires a European counterparty under European law, you need one of the tools below, because no setting inside LangSmith changes the entity on the order form.
Can a European tool read traces that my LangChain code already emits?
Possibly. LangWatch, Latitude, Orq.ai, Rhesis AI and Coralogix list OpenTelemetry support or OpenTelemetry ingestion on their vendor pages, but we did not verify how LangChain code maps onto each of them. The exact mapping of span names and attributes differs per tool, so run a pilot service for a week and compare what shows up.
Evaluations do not move in the same way. A LangSmith evaluator is code that reads a run and an example. Most alternatives let you rebuild it as a judge prompt, a Python metric or a pytest check, which is fast for simple scorers and slower for ones that depend on LangSmith dataset features.
Which one to pick
If you want the most direct replacement, start a pilot on LangWatch, and add Latitude if your pain is recurring production failures rather than missing tests. Pick Galtea, Giskard or Rhesis AI when the main job is pre-release testing, with Giskard for security probes, Galtea for generated test sets and Rhesis AI for expert review. These can sit next to a tracing tool instead of replacing it.
Choose Orq.ai or Opper if the real need is controlled model access with logs attached, Coralogix if an observability platform is already on the contract, and Qualifire if blocking output at request time matters most. Stay on LangSmith if you rely on its deployment, Fleet or sandbox products, or if an EU-hosted instance with a US contracting party passes your review. The honest answer is that it is the broadest product here.
Frequently Asked Questions
For tracing and evaluations in one product, LangWatch from Amsterdam is the closest match, with Apache-2.0 code and an EU cloud region. Latitude from Barcelona suits production teams that want failing traces grouped into issues. Galtea and Giskard focus on pre-release testing, Orq.ai and Opper add a gateway, and Coralogix fits companies that already run it for infrastructure.
The regions FAQ says LangSmith complies with the GDPR, is SOC 2 Type 2 certified and HIPAA compliant, and that a DPA can be signed through support. Compliance does not change the contracting party. LangChain states it has no EU legal entity for customers, so a data processing agreement is made with a Delaware corporation even when the EU region is selected.
The pricing page lists self-hosted and hybrid deployment under the Enterprise plan at custom pricing, and the terms define self-hosted, hybrid and bring-your-own-cloud variants. In the hybrid variant, traces from customer workloads go to LangChain cloud. LangWatch, Latitude, Rhesis AI and ZenML publish open-source code that you can run yourself without an enterprise contract.
LangSmith has a free Developer seat with 5k base traces, then Plus at $39 per seat per month with 10k base traces, plus pay-as-you-go units at $1.00 each. LangWatch Growth is EUR 29 per core seat, Latitude Pro is $99 a month for the workspace, and Opper and Orq.ai charge a percentage of model spend. Compare them on your own volume.
Giskard from Paris generates adversarial and quality tests and has an Apache-2.0 library. Galtea from Barcelona creates test cases from specifications and simulates conversations, and LangWatch runs simulated users inside pytest or vitest. Rhesis AI from Potsdam adds expert review, so non-engineers can judge answers. Pick based on who writes the pass criteria on your team.
Related comparisons
About the author
Explore More European Alternatives
Discover privacy-focused European alternatives to other popular US tech services.