LangSmith alternatives for tracing and evaluating production agents
The main LangSmith alternatives in 2026 are Langfuse and Comet Opik for open-source self-hosting, Braintrust for eval-first teams, Arize Phoenix for OpenTelemetry-native tracing, Pydantic Logfire for full-stack Python observability, and Datadog, W&B Weave or MLflow for teams already on those platforms.
Checked
LangSmith is LangChain's hosted platform for tracing, evals, datasets and prompts. It works outside LangChain, accepts OpenTelemetry traces, and added LangSmith Engine in May 2026 to cluster production failures and generate evaluators. Teams still look elsewhere, usually for one of three reasons: self-hosting is Enterprise only, pricing is per seat, or the team is standardised on another platform.
The alternatives below are grouped by what they are strongest at. Each one fits a different kind of team running agents in production; none is a drop-in replacement for every LangSmith feature.
Why teams look past LangSmith
Self-hosting
Self-hosted and hybrid LangSmith deployments are on the Enterprise plan only. Teams with data-residency rules often need a tool they can run themselves without a custom contract.
Pricing shape
LangSmith Plus is $39 a seat a month with 10k base traces, and base traces are kept 14 days. Usage-priced tools can cost less for large teams with modest trace volume.
Existing platforms
An organisation already paying for Datadog, Databricks or Weights & Biases may prefer agent traces next to the rest of its telemetry.
Open-source and self-hosted options
Langfuse
MIT core, owned by ClickHouse since 16 January 2026. The open-source edition is free to self-host without limits on core features, and covers tracing, prompts, datasets, LLM-as-a-judge and code evaluators.
Comet Opik
Apache-2.0, with a self-hosted version on the same codebase as the hosted one. Opik Optimizer 3.0.0 (January 2026) adds automated prompt and agent optimisation.
Arize Phoenix
Built on OpenTelemetry and OpenInference and deployable with Docker or a Helm chart. Its licence is the Elastic License 2.0, which is source-available rather than OSI open source.
MLflow 3
Apache-2.0, with tracing, mlflow.genai.evaluate() scorers and LLM judges, and production monitoring.
Hosted and commercial options
Braintrust
Eval-first: experiments, scorers and CI joined to production traces, with Topics (GA June 2026) clustering failures. Starter is $0, Pro is $249 a month, and self-hosting is Enterprise only.
Pydantic Logfire
OpenTelemetry-native observability for the whole app, from database and HTTP calls to model calls, with online evals since 30 April 2026. Team is $49 a month and Enterprise can self-host via Helm.
Datadog Agent Observability
Formerly LLM Observability, renamed in 2026. Free up to 40k LLM spans a month; Pro from $160 a month on annual terms. Only LLM-call spans are billed.
W&B Weave
Now part of CoreWeave Forge. Free for 1 GB of ingestion a month, Pro from $60 a month. Fits teams that already track training runs in W&B.
What changed in 2026
ClickHouse bought Langfuse in January. Mintlify bought Helicone in March and put it into maintenance mode, which rules it out for new builds. Galileo became Splunk Agent Observability after Cisco's acquisition. W&B accounts started signing in through CoreWeave Forge on 30 September. The OpenAI Evals platform went read-only on 31 October and shuts down on 30 November 2026, with Promptfoo as OpenAI's named migration path.
How to choose
Start from constraints, not features: where traces may be stored, which licence the legal team accepts, and which platform the organisation already pays for. Then run a short trial on real agent traces, checking that tool calls nest correctly, costs match the provider's invoice and evaluators catch known failures. Instrumenting with OpenTelemetry keeps a later switch cheaper, though the GenAI conventions are still at Development status.
Motiontic's engineers are vendor-neutral on tracing and eval tools. They pick a backend against the client's hosting, data-residency and budget constraints, keep instrumentation portable where the stack allows, build the eval suite from the client's own production failures, and leave the traces, datasets and evals with the client's team.
LangSmith alternatives for production agents. Prices in USD, as of 1 November 2026
| Licence / hosting | Strength | Fits when | |
|---|---|---|---|
| Langfuse | MIT core (ClickHouse); free self-host or cloud from $0 | Tracing, prompts, datasets, LLM-as-a-judge and code evaluators | Traces must stay on the team's own infrastructure |
| Braintrust | Proprietary, SDK Apache-2.0; self-host on Enterprise; Pro $249 a month | Experiments, scorers and CI joined to production traces | Evals are the release gate |
| Arize Phoenix | Elastic License 2.0 (source-available); Docker or Helm | Built on OpenTelemetry and OpenInference; datasets, experiments | The team wants OTel-native tracing it runs itself |
| Comet Opik | Apache-2.0; self-host or cloud, Pro $19 a month for 100k spans | Tracing, evals and an agent and prompt optimizer | A permissive OSI licence matters |
| Pydantic Logfire | SDK MIT; Team $49 a month; Enterprise self-host via Helm | Full-stack OpenTelemetry plus online evals | A Python or Pydantic AI team wants one tool for app and agent traces |
| Datadog Agent Observability | Proprietary, inside Datadog; free to 40k LLM spans a month | Agent traces beside APM, infrastructure and logs | The organisation already runs on Datadog |
| W&B Weave | SDK Apache-2.0; CoreWeave Forge, Pro from $60 a month | Training runs and app-level evals in one place | The team already tracks training in W&B |
| MLflow 3 | Apache-2.0 | Tracing, scorers, LLM judges and production monitoring | The team runs on Databricks or already uses MLflow |
Common questions
What is the closest open-source alternative to LangSmith?
Langfuse covers the same ground of tracing, prompts, datasets and evals under an MIT core and can be self-hosted free. Comet Opik is the other full platform under a permissive licence, Apache-2.0.
Is there a free alternative to LangSmith?
Several. Langfuse, Opik, MLflow and Arize Phoenix can be self-hosted at no licence cost, and Langfuse, Opik, Braintrust, Logfire and Datadog all have free hosted tiers with usage limits.
Which LangSmith alternatives accept OpenTelemetry traces?
Langfuse and Braintrust accept OTLP traces, Phoenix is built on OpenTelemetry, and Logfire is OpenTelemetry-native. LangSmith itself also accepts OTLP over HTTP, which makes moving between these tools easier.
Is Helicone still a LangSmith alternative?
Not for new builds. Mintlify acquired Helicone on 3 March 2026 and put it into maintenance mode, limited to security fixes, bug fixes and new models, and offered to help customers migrate.
Related
AI agent development
Agents that do real work for users, with the evals, traces and guardrails to keep them in production.
LangSmith vs Langfuse
Hosted platform from LangChain against the open-source, self-hostable platform now owned by ClickHouse.
Braintrust vs LangSmith
Two hosted platforms for agent traces and evals, one built around experiments, one around the LangChain stack.
Langfuse alternatives
Seven tracing and eval platforms compared on licence, hosting, strengths and the teams each one fits.
OpenTelemetry GenAI semantic conventions
Standard gen_ai.* names for agent and model telemetry, still at Development status.
Sources
- LangChain: LangSmith pricing
- Langfuse: Self-host pricing
- Braintrust: Pricing
- GitHub: Arize-ai/phoenix
- Comet: Opik pricing
- Pydantic: Logfire pricing
- Datadog: Agent Observability pricing
- CoreWeave Forge pricing (W&B Weave)
- MLflow 3 for GenAI
- Mintlify acquires Helicone (2026-03-03)
- Splunk: Galileo acquisition
- OpenAI: API deprecations
Running agents that have to hold up?
Describe the agent, the systems it touches and what it must never get wrong. The first call is with a lab engineer, who will say plainly whether it is ready for production, and whether Motiontic is the right team.