LLM observability tools compared (2026)
If you need an EU data region, Braintrust, Langfuse and LangSmith state one on their official pages. If you need to keep traces in your own infrastructure, Helicone, Laminar, Langfuse and Opik (Comet) are self-hostable under MIT or Apache-2.0 per their official pages. Every cell below comes from the vendor's own site, docs or GitHub repository, and "not stated" means we found no statement, not that the feature is missing.
Last checked 2026-10-30. Tools are listed alphabetically. This is not a ranking and we did not test the tools ourselves.
Comparison table
| Tool | Open source / license | Self-host | EU data region | Free plan | OpenTelemetry |
|---|---|---|---|---|---|
| Arize Phoenix | ELv2 (Elastic License 2.0) | Yes | Not stated | Free tier stated ("start free", no credit card required); no limits stated | Yes |
| Braintrust | Not stated | Not stated | Yes | Starter, $0/month: $10 credits, 1 GB processed data, 10k scores, 14-day retention, unlimited users (checked 2026-10-30) | Yes |
| Datadog LLM Observability | Not stated | Not stated | Not stated | Free plan, $0/month: up to 40k LLM spans, 15-day retention (checked 2026-10-30) | Yes |
| Helicone | Apache-2.0 | Yes | Not stated | Hobby, $0: 10,000 free requests, 1 GB storage, 7-day retention, 1 seat (checked 2026-10-30) | Not stated |
| Laminar | Apache-2.0 | Yes | Not stated | Free, $0/month: 1 GB data, 7-day data access, 1 project, 1 seat (checked 2026-10-30) | Yes |
| Langfuse | MIT, except the ee folders | Yes | Yes | Hobby, free, no credit card: 50k units per month, 30 days of data access, 2 users (checked 2026-10-30) | Yes |
| LangSmith | Not stated | Yes | Yes | Developer, $0 per seat per month: 1 seat, 5k base traces per month, then pay as you go (checked 2026-10-30) | Yes |
| Langtrace | AGPL-3.0 (application), Apache-2.0 (SDKs) | Yes | Not stated | Starter, "Free forever": up to 10K spans per month, 1 user (checked 2026-10-30) | Yes |
| Lunary | Not stated | Yes | Not stated | Not stated | Yes |
| Opik (Comet) | Apache-2.0 | Yes | Not stated | Free Cloud plan, $0: up to 10 team members, 25k spans per month, 60-day retention (checked 2026-10-30) | Yes |
| Traceloop / OpenLLMetry | Apache-2.0 (OpenLLMetry) | Yes | Not stated | Not stated | Yes |
| W&B Weave | Apache-2.0 | Not stated | Not stated | Not stated | Yes |
"Yes" for EU data region means the vendor states an EU option for its cloud. It does not mean every plan includes it. See each card for the wording and sources.
Tool by tool
Arize Phoenix
Open-source AI observability and evaluation: tracing, evals and dataset-driven experiments for LLM apps and agents.
Suits: Teams that want an OpenTelemetry-native tool they can run in their own environment with Docker or Kubernetes.
Notes: ELv2 is source-available with restrictions, not an OSI open-source license. Read it before offering Phoenix as a hosted service.
Sources: arize.com/phoenix, github.com/Arize-ai/phoenix
Braintrust
Observability platform for agents: trace runs, score outputs with evals and review production patterns.
Suits: Teams that put evals at the center and want a managed service with an EU API region or a hybrid data plane in their own infrastructure.
Notes: Not listed as self-host. The homepage describes a hybrid option where the Brainstore data plane runs on your own infrastructure, and the pricing page puts on-prem or hosted deployment under Enterprise. The platform's open-source status is not stated; the site links to separate open-source libraries.
Sources: braintrust.dev, braintrust.dev/pricing, braintrust.dev/docs/kb/browser-and-api-dom, braintrust.dev/docs/integrations/sdk-integ
Datadog LLM Observability
Datadog's agent and LLM tracing, experiments and production monitoring, inside the Datadog platform.
Suits: Teams that already run Datadog and want LLM traces next to their existing APM and logs.
Notes: Datadog documents an EU1 site (EU, Germany) in general. The product page and LLM Observability docs fetched do not say whether LLM Observability is available there, so confirm with Datadog before relying on it.
Sources: datadoghq.com/product/llm-observability/, docs.datadoghq.com/llm_observability/instrumen, docs.datadoghq.com/getting_started/site/
Helicone
AI gateway and LLM observability: route, log, debug and analyze LLM requests.
Suits: Teams that want a proxy-style gateway with logging, self-hosted with Docker Compose.
Notes: The GitHub README documents Docker Compose self-hosting; the pricing page lists on-prem deployment under Enterprise. OpenTelemetry is not stated; the README only lists OpenLLMetry in an integrations table.
Sources: helicone.ai, helicone.ai/pricing, github.com/Helicone/helicone
Laminar
Open-source agent observability platform that captures LLM calls, tool calls, sub-agents, costs and tokens.
Suits: Teams building agents who want a permissively licensed, self-hostable backend (Docker Compose or Helm).
Notes: www.lmnr.ai redirects to laminar.sh. The GitHub README calls the tracing SDK OpenTelemetry-native.
Sources: laminar.sh, laminar.sh/pricing, github.com/lmnr-ai/lmnr
Langfuse
Open-source observability and engineering platform for LLM apps and agents: tracing, prompt management, evaluation and metrics.
Suits: EU teams that want one tool with an EU cloud region and a free self-hosted option.
Notes: The pricing page says cloud data can be stored in the US, EU or Japan, including on the Hobby plan.
LangSmith
LangChain's agent observability and evaluation platform: tracing, monitoring, insights and evals.
Suits: Teams already building on LangChain or LangGraph.
Notes: The pricing page lists cloud data location as "LangChain's Cloud (US or EU)" and does not say which plans include the EU option. The product page FAQ says LangSmith offers managed cloud, bring-your-own-cloud and self-hosted options for teams with data residency requirements; the pricing page lists self-hosted and hybrid under Enterprise. Open-source status of LangSmith itself is not stated.
Sources: langchain.com/langsmith, langchain.com/pricing
Langtrace
Open-source observability tool for LLM apps: tracing, evaluations and metrics, with Python and TypeScript SDKs.
Suits: Small teams that want a self-hosted stack (Next.js app, Postgres, ClickHouse) and can accept AGPL terms.
Notes: The www.langtrace.ai homepage returned a 404 when checked, so this entry uses the official docs and GitHub repository.
Sources: docs.langtrace.ai/introduction, github.com/Scale3-Labs/langtrace
Lunary
Platform for chatbots and LLM apps: agent traces, cost and usage analytics, prompt versioning, PII masking, RBAC and SSO.
Suits: Teams that want to deploy in their own VPC with Docker or Kubernetes and need chatbot analytics.
Notes: The homepage lists a cloud Team workspace at $20 per user per month, says payment is required to activate a workspace, and does not mention a free plan. The docs say Lunary ingests OTEL traces through a /v1/otel endpoint. No license is stated on the pages checked.
Sources: lunary.ai, docs.lunary.ai/llms.txt, docs.lunary.ai/integrations/opentelemetry/
Opik (Comet)
Open-source platform for tracing, evaluating and monitoring LLM applications, RAG systems and agent workflows.
Suits: Teams that want an Apache-2.0 tool with the same codebase in the hosted and self-hosted versions.
Notes: The pricing page lists the data region as US for the Free and Pro plans and "Custom" for Enterprise, so EU hosting is not stated for the cloud plans.
Sources: comet.com/site/products/opik/, comet.com/site/pricing/, github.com/comet-ml/opik
Traceloop / OpenLLMetry
Evals and monitoring for LLM apps, built on OpenLLMetry, an open-source OpenTelemetry-based SDK.
Suits: Teams that want OpenTelemetry-standard instrumentation they can send to any OTel backend.
Notes: The homepage says Traceloop can be deployed in the cloud, on-prem or air-gapped. The OpenLLMetry README does not mention self-hosting a backend. The homepage also announces that Traceloop is joining ServiceNow.
W&B Weave
Observability and evaluation toolkit from Weights & Biases for tracing and evaluating agents and LLM applications.
Suits: Teams already using Weights & Biases who want traces in the same account.
Notes: The Weave docs now live on docs.coreweave.com and require a CoreWeave Forge account and API key. They mention dedicated Weave instances but do not describe self-hosting, EU hosting or a free tier.
Sources: wandb.ai/site/weave, github.com/wandb/weave, docs.coreweave.com/weave/
How to choose if your data must stay in the EU
1. Decide whether a cloud EU region is enough
An EU region keeps stored traces in the EU, but your prompts, outputs and user identifiers may still contain personal data. Check the vendor's data processing agreement and sub-processor list before you send production traffic. Among the tools here, Braintrust, Langfuse and LangSmith state an EU cloud option. Langfuse says cloud data can be stored in the US, EU or Japan, and that this applies to its free Hobby plan too.
2. Self-host when the region question is not enough
Self-hosting moves the question from "which region does the vendor use" to "where do we run it". It costs you operations work: Langtrace, for example, needs a Next.js app, Postgres and ClickHouse. Read the license first. Arize Phoenix is ELv2 (Elastic License 2.0), and Langtrace's application is AGPL-3.0, which can matter for commercial use.
3. Keep your instrumentation portable with OpenTelemetry
Vendor-specific tracing code ties you to one backend. Tools that state OpenTelemetry support (Arize Phoenix, Braintrust, Datadog LLM Observability, Laminar, Langfuse, LangSmith, Langtrace, Lunary, Opik (Comet), Traceloop / OpenLLMetry and W&B Weave) accept standard traces, which keeps the option of switching backends. Where OpenTelemetry is not stated, ask how traces are exported.
4. Check what the free plan really covers
Free plans differ in retention (7 to 60 days among those that state it), users and volume. Prices and limits were checked on 2026-10-30 and can change, so use the pricing pages in each card.
Choosing a model provider too? See our pages on OpenAI and Claude, or the wider AI development overview.
Need help setting it up?
If you would rather hand setup to a specialist, browse machine learning development agencies, see the agencies that work with LangChain, or get matched with an agency. Looking by location? See agencies in France, Poland and the Netherlands. Our listings do not say which observability tool an agency uses, so ask about it in your brief.
Sources
All fetched on 2026-10-30.
- https://arize.com/phoenix
- https://github.com/Arize-ai/phoenix
- https://www.braintrust.dev
- https://www.braintrust.dev/pricing
- https://www.braintrust.dev/docs/kb/browser-and-api-domains-for-braintrust-orgs.md
- https://www.braintrust.dev/docs/integrations/sdk-integrations/opentelemetry/send-traces-and-logs.md
- https://www.datadoghq.com/product/llm-observability/
- https://docs.datadoghq.com/llm_observability/instrumentation/otel_instrumentation/
- https://docs.datadoghq.com/getting_started/site/
- https://www.helicone.ai
- https://www.helicone.ai/pricing
- https://github.com/Helicone/helicone
- https://laminar.sh/
- https://laminar.sh/pricing
- https://github.com/lmnr-ai/lmnr
- https://langfuse.com/pricing
- https://github.com/langfuse/langfuse
- https://www.langchain.com/langsmith
- https://www.langchain.com/pricing
- https://docs.langtrace.ai/introduction
- https://github.com/Scale3-Labs/langtrace
- https://lunary.ai
- https://docs.lunary.ai/llms.txt
- https://docs.lunary.ai/integrations/opentelemetry/overview.md
- https://www.comet.com/site/products/opik/
- https://www.comet.com/site/pricing/
- https://github.com/comet-ml/opik
- https://www.traceloop.com
- https://github.com/traceloop/openllmetry
- https://wandb.ai/site/weave
- https://github.com/wandb/weave
- https://docs.coreweave.com/weave/
Questions
Which LLM observability tools state an EU data region?
On the official pages we read, Braintrust, Langfuse and LangSmith state an EU option. For the other 9 tools in the table, an EU region is not stated, which is different from "not available": ask the vendor before you assume either way. Datadog documents an EU1 site (EU, Germany) in general. The product page and LLM Observability docs fetched do not say whether LLM Observability is available there, so confirm with Datadog before relying on it.
Which tools can I self-host?
Arize Phoenix, Helicone, Laminar, Langfuse, LangSmith, Langtrace, Lunary, Opik (Comet) and Traceloop / OpenLLMetry have self-hosting stated on their official site, docs or GitHub repository. Of these, Helicone, Laminar, Langfuse and Opik (Comet) are under MIT or Apache-2.0 according to the pages we checked. Self-hosting keeps traces in your own infrastructure, so data residency becomes a question about where you run it.
Is Arize Phoenix open source?
Phoenix's repository states ELv2 (Elastic License 2.0). That license is source-available with restrictions, so read it before you build a hosted service on top of it. Phoenix can be self-hosted with Docker or Kubernetes.
Do these tools support OpenTelemetry?
Arize Phoenix, Braintrust, Datadog LLM Observability, Laminar, Langfuse, LangSmith, Langtrace, Lunary, Opik (Comet), Traceloop / OpenLLMetry and W&B Weave state OpenTelemetry support in their docs or repositories. Helicone did not state it on the pages we checked. OpenTelemetry-based instrumentation makes it easier to switch backends later.
Which tools have a free plan?
Several publish one, including Braintrust, Datadog LLM Observability, Helicone, Laminar, Langfuse, LangSmith, Langtrace and Opik. Limits are in the table and were checked on 2026-10-30. Prices and limits change, so confirm on each pricing page.
Does GDPR require an EU region?
This page does not give legal advice. An EU region keeps stored traces in the EU, but you still need to look at what personal data your prompts and responses contain, who the vendor's sub-processors are, and what a data processing agreement covers. Ask your data protection officer.
Get matched with 3 vetted agencies
Describe the project. We read every brief and introduce you to three agencies from this index that fit it, usually within 2 working days. Free for buyers; agencies can't pay to be picked.