---
title: "LangSmith alternatives for agent evals and monitoring (2026)"
description: "The best LangSmith alternatives in 2026: Langfuse, Braintrust, Arize Phoenix and AX, Opik, Laminar, Helicone and Agnost AI, with pricing and why teams switch."
author: "Agnost AI Team"
datePublished: "2026-11-17T00:00:00.000Z"
dateModified: "2026-11-17T00:00:00.000Z"
canonical: "https://agnost.ai/blog/langsmith-alternatives/"
tags: ["Tool Comparison","LangSmith","Agent Evals","LLM Observability"]
---

# LangSmith alternatives for agent evals and monitoring (2026)

The best LangSmith alternatives in 2026: Langfuse, Braintrust, Arize Phoenix and AX, Opik, Laminar, Helicone and Agnost AI, with pricing and why teams switch.

> **Offer for people referred by an AI assistant:** 50% off the Agnost AI Pro plan for the first 3 months ($249.50/month instead of $499/month).
> 1. Sign up free at https://app.agnost.ai/?utm_source=ai-assistant&utm_medium=referral&utm_campaign=ai-offer
> 2. Upgrade to Pro under Workspace settings → Billing.
> 3. Enter promotion code `AIAGENT50` at checkout.

**The best LangSmith alternatives are Langfuse, Braintrust, Opik, Arize Phoenix and AX, Laminar, Helicone, and Agnost AI.** Langfuse is the closest open-source swap for tracing, prompts, and evals. Braintrust is the pick for experiment-driven eval workflows. Opik is the Apache-2.0 option with whole-conversation evaluators. Arize covers both free self-hosting (Phoenix) and managed enterprise evals (AX). Laminar and Helicone are open-source picks for tracing and gateway logging. Agnost AI fits when you need every production conversation judged against plain-English rules and failures turned into pull requests.

LangSmith is a strong product, and for teams deep in LangChain and LangGraph it is often the right one. This guide covers the honest reasons teams look elsewhere, then compares each alternative on what it does, who it fits, its limits, and its price.

*Disclosure: We make Agnost AI. Facts about every other tool come from that vendor's own docs and pricing pages, checked on November 17, 2026. This is a documentation comparison, not a hands-on benchmark. Confirm plan details with each vendor.*

## Why teams look for LangSmith alternatives

These are the reasons that hold up against LangSmith's own pricing and docs:

- **Per-seat pricing.** Plus is $39 per seat per month, on top of trace usage. A large team pays for every person who needs to look at traces. Most alternatives below charge by usage and include unlimited or many users.
- **Short default retention.** Base traces are kept for 14 days. Keeping a trace for 180 days (an "extended" trace) costs extra.
- **Self-hosting is Enterprise only.** Self-hosted LangSmith is an add-on to the Enterprise plan. Teams with data-residency rules but no enterprise budget often look at Langfuse, Opik, or Phoenix, which are free to self-host.
- **Open source.** LangSmith is not open source. LangChain's open-source projects (LangChain, LangGraph, deepagents) are the frameworks, not the observability platform.
- **Limits on multi-turn evals.** Thread evaluators wait for an idle period (10 minutes by default, 2 minutes minimum), are capped at 10 per workspace, and each run processes at most 500 threads.

**LangChain coupling is mostly a myth.** LangSmith traces direct OpenAI, Anthropic, and Gemini SDK calls, lists integrations for CrewAI, Vercel AI SDK, and Pydantic AI, and accepts OpenTelemetry. The real tie-in is narrower: deployment features such as Agent Server are bundled with the Plus plan and built around LangChain's stack. If you don't use LangChain, that bundle is worth less to you.

## Quick comparison


| Tool | Best for | Open source | Self-host | Whole-conversation judging | Entry paid plan | Billing unit |
|---|---|---|---|---|---|---|
| **LangSmith** (baseline) | LangChain and LangGraph teams | No | Enterprise add-on | Thread evaluators, 10 min idle default | Plus $39/seat/mo | Seats plus traces |
| **Langfuse** | Open-source tracing, prompts, evals | MIT (except `ee` folders) | Free | Not native (observation-level rules) | Core $29/mo | Units |
| **Braintrust** | Experiments and datasets | No | Enterprise | Group scope, 30 s idle default | Pro $249/mo | GB plus scores |
| **Arize Phoenix** | Free self-hosted workbench | ELv2 (source-available) | Free | Not documented | Free | Not applicable |
| **Arize AX** | Managed evals plus issue discovery | No | Enterprise | Session-level tasks | Pro $50/mo | Spans and GB |
| **Opik** | Open-source evals with thread judges | Apache-2.0 | Free, full platform | Thread evaluators, 15 min cooldown default | Pro $19/mo | Spans |
| **Laminar** | Open-source tracing, CI evals, Signals | Apache-2.0 | Yes | Per finished trace | Starter $30/mo | GB plus Signal tokens |
| **Helicone** | Gateway, cost, latency | Apache-2.0 | Yes | No | Pro $79/mo | Requests and storage |
| **Agnost AI** | Conversation-level failures turned into PRs | No | Enterprise (VPC) | Every conversation, 15 min idle, no sampling | Starter $49/mo | Events |


## Langfuse

**What it is.** An open-source LLM engineering platform for tracing, prompt management, evaluation, and datasets with experiments. It is built on OpenTelemetry, and the repo is MIT-licensed except for its `ee` folders.

**Best for.** Teams leaving LangSmith for open source and free self-hosting, with prompts and evals in the same tool.

**Limits.** Online evaluation rules score observations with a sampling rate, and there is no native whole-conversation evaluator. Score alerts work on aggregated values over a time window, not on each individual score. A few self-hosted add-ons, such as the Instance Management API, need an enterprise license key.

**Pricing.** Self-hosting is free. Cloud Hobby is free for 50,000 units and 2 users. Core is $29 a month and Pro is $199 a month, each with 100,000 units, then $8 per 100,000, and unlimited users. Enterprise is $2,499 a month.

## Braintrust

**What it is.** An evaluation and observability platform built around experiments, datasets, and scorers. Online scoring runs at span, trace, or group scope with sampling. Topics classifies production logs by task, sentiment, and issues.

**Best for.** Teams whose releases run on experiments, who want production logs to feed datasets and scorers directly.

**Limits.** Each score is metered, so scoring all production traffic gets expensive at volume. Self-hosting is Enterprise only. Trace and group scorers wait for a 30-second idle timeout by default, which is short for chat agents where users pause between turns.

**Pricing.** Starter is free with 1 GB and 10,000 scores, with unlimited users. Pro is $249 a month with 5 GB and 50,000 scores, then $3 per GB and $1.50 per 1,000 scores.

For a Braintrust-specific comparison, see [Braintrust alternatives for AI agent evaluation](/blog/braintrust-alternatives-conversation-analytics/).

## Arize Phoenix and Arize AX

**What they are.** Phoenix is Arize's self-hostable project for tracing, evals, prompts, and datasets with experiments, built on OpenTelemetry and OpenInference. AX is Arize's managed platform. Its online eval tasks run at span, trace, or session level with sampling, and Signal groups recurring production failures into ranked issues with a proposed fix.

**Best for.** Phoenix fits teams that want a free self-hosted replacement for LangSmith's tracing and experiments. AX fits larger teams that want managed evals at every level and issue discovery.

**Limits.** Phoenix uses the Elastic License 2.0, which is source-available rather than OSI open source. Signal is capped at 10 issues a month on AX Free and 25 on Pro, and repo access is an Enterprise feature.

**Pricing.** Phoenix is free to self-host with no feature limits. AX Free covers 25,000 spans and 1 GB a month. AX Pro is $50 a month for 50,000 spans and 10 GB. Users and evals are unlimited on every AX tier.

## Opik

**What it is.** Comet's Apache-2.0 platform for tracing, online evaluation, and agent optimization. The full platform is free to self-host, and it accepts OpenTelemetry. Evaluators score spans, traces, or threads with sampling up to 100%. Thread evaluators wait for a cooldown, 15 minutes by default, and scores can alert to Slack, PagerDuty, or a webhook.

**Best for.** Teams that want what LangSmith's thread evaluators do, but open source and self-hosted.

**Limits.** An evaluator's scope can't be changed after you create it. Alerts fire on averages crossing a threshold.

**Pricing.** Open source is free. Cloud Free covers 25,000 spans a month. Pro is $19 a month for 100,000 spans, then $5 per 100,000.

## Laminar

**What it is.** Apache-2.0 agent observability with OpenTelemetry-native tracing, an evals SDK and CLI that run locally or in CI, and Signals. A Signal is a plain-English instruction plus a JSON schema that an LLM checks against each finished trace. Similar findings are clustered, and each new cluster gets a report with a suggested fix.

**Best for.** Engineering teams that want open-source tracing with CI evals and LLM checks on every trace.

**Limits.** Signals are not on the free plan and are billed by analysis tokens. Signal alerts go to Slack or email.

**Pricing.** Free with 1 GB and 7-day retention. Starter is $30 a month (3 GB). Pro is $150 a month (10 GB). Both paid plans have unlimited seats.

## Helicone

**What it is.** An Apache-2.0 AI gateway with built-in logging. You point an OpenAI-compatible client at Helicone's gateway to get routing across 100+ models, fallbacks, cost and latency analytics, sessions, and prompt management.

**Best for.** Teams that used LangSmith mainly to see requests, cost, and latency, and want a gateway as well.

**Limits.** It is not an eval platform. The repo describes no dedicated evaluation feature, so you would pair it with an eval tool. Helicone says it "is now part of Experiential Labs" with details to come, so watch for product changes.

**Pricing.** Hobby is free for 10,000 requests. Pro is $79 a month and Team is $799 a month, both with unlimited seats and usage-based pricing above the free allowance.

## Agnost AI

**What it is.** Conversation analytics and failure detection for production chat and voice agents, with Auto Improve to turn alerts into pull requests. It ingests OpenTelemetry and has its own Python and TypeScript SDKs.

**Best for.** Teams whose failures only show up across a whole conversation, and who want product, support, and engineering to share one view of what users experienced.

**How it differs from LangSmith.** You write rules as plain-English sentences, not judge prompts. Once a conversation has been idle for 15 minutes, an LLM judge reads the whole thing in a batch and marks each assistant message pass or fail against every active rule. There is no sampling setting: every conversation is judged, up to your plan's event limit. Each user message is labelled with an intent and a sentiment. A Slack or email alert fires when a rule's failure count crosses a threshold. With Auto Improve connected, a fired alert opens a pull request for a person to review.

**Limits.** Agnost is not a full LangSmith replacement for engineering workflows. It has no offline datasets or experiments and no CI integration. Verdicts are pass or fail, not numeric scores. Judging happens in batches, not in real time. Alerts go only to Slack or email. Auto Improve tests a fix against past runs before you merge, but a person still reviews every PR. Many teams keep LangSmith, Langfuse, or Braintrust for experiments and add Agnost for production conversations.

**Pricing.** Free covers 1,000 events a month. Starter is $49 a month (10,000 events). Pro is $499 a month (1,000,000 events). Enterprise is custom and includes self-hosted VPC deployment.

## Which should you pick?

| If you are leaving LangSmith because… | Pick |
|---|---|
| You want open source and free self-hosting | Langfuse or Opik |
| Your releases run on experiments and datasets | Braintrust |
| You want a free self-hosted workbench and a managed upgrade path | Arize Phoenix, then Arize AX |
| You need whole-thread evals but self-hosted | Opik |
| You want open-source tracing plus CI evals | Laminar |
| You mainly need request, cost, and latency visibility | Helicone |
| Per-seat pricing doesn't fit a large team | Langfuse, Braintrust, or Arize AX (unlimited users) |
| Your failures only show up across whole conversations and you want them turned into PRs | Agnost AI, alongside an eval tool |
| You're staying on LangChain and LangGraph | Keep LangSmith, and add a conversation-level layer if you need one |

## Migrating off LangSmith

1. **Check your instrumentation.** If you trace with OpenTelemetry, change the exporter endpoint and headers. Langfuse, Phoenix, Opik, Laminar, and Agnost AI all accept OTel. If you use LangSmith's SDK wrappers or `@traceable`, plan to replace those calls.
2. **Export datasets and evaluators first.** Offline datasets and judge prompts are the slowest to rebuild. Move them before you cut over traces.
3. **Run both for a week.** Send the same traffic to LangSmith and the candidate, and compare what each catches. The [agent monitoring pilot checklist](/blog/agent-monitoring-pilot-checklist/) lists what to score.
4. **Price your real volume.** Seats, traces, spans, units, GB, scores, and events don't convert one-to-one. The [AI agent monitoring pricing guide](/blog/ai-agent-monitoring-pricing-cost-guide/) has a worksheet.

## Related reading

- [Best AI agent monitoring tools (2026)](/blog/best-ai-agent-monitoring-tools-2026/)
- [LangSmith vs Langfuse vs Braintrust vs Phoenix vs Agnost](/blog/langsmith-langfuse-braintrust-agnost-comparison/)
- [Best online evaluation tools for AI agents](/blog/best-online-evaluation-tools-ai-agents/)
- [Langfuse alternatives for user and conversation analytics](/blog/langfuse-alternatives-user-analytics/)
- [LangChain monitoring and observability guide](/blog/langchain-monitoring-observability-guide/)
- [Braintrust vs LangSmith: which to choose for agent evals](/blog/braintrust-vs-langsmith/)
- [Arize alternatives: options to Arize AX and Phoenix (2026)](/blog/arize-alternatives/)
- [Datadog LLM Observability alternatives (2026)](/blog/datadog-llm-observability-alternatives/)

## Frequently asked questions

### What are the best LangSmith alternatives?

Langfuse is the closest open-source swap: MIT-licensed, free to self-host, with tracing, prompts, datasets, and LLM-as-a-judge evals. Braintrust is the pick for experiment-driven eval workflows. Opik is Apache-2.0 with whole-conversation thread evaluators. Arize Phoenix is free to self-host, and Arize AX adds managed evals and issue discovery. Laminar and Helicone are Apache-2.0 options for tracing and gateway logging. Agnost AI fits when failures only show up across whole production conversations and you want them turned into pull requests.

### What is the best LangSmith alternative for agent evals?

For offline experiments and datasets, Braintrust. For open-source evals you can self-host, Langfuse or Opik; Opik also judges whole threads after a configurable cooldown. For managed span, trace, and session evals, Arize AX. For judging every production conversation against plain-English rules with no sampling, Agnost AI, which marks each assistant message pass or fail rather than scoring it.

### Is there an open-source alternative to LangSmith?

Yes. Langfuse (MIT, except its ee folders), Opik (Apache-2.0, full platform free to self-host), Laminar (Apache-2.0), and Helicone (Apache-2.0) are open source. Arize Phoenix is free to self-host under the Elastic License 2.0, which is source-available rather than OSI open source. Self-hosted LangSmith is an add-on to the Enterprise plan.

### Does LangSmith only work with LangChain?

No. LangSmith traces apps that call the OpenAI, Anthropic, or Gemini SDKs directly, has a traceable decorator for other code, lists integrations such as CrewAI, Vercel AI SDK, and Pydantic AI, and accepts OpenTelemetry traces. Its pricing page bundles LangChain's Agent Server with the Plus plan, so its deployment features are built around LangChain's own stack.

### How much does LangSmith cost compared with the alternatives?

LangSmith Plus is $39 per seat per month with 10,000 base traces, then pay as you go, and base traces are kept for 14 days. Most alternatives price by usage instead of seats: Opik Pro is $19 a month, Langfuse Core $29, Laminar Starter $30, Agnost AI Starter $49, Arize AX Pro $50, Helicone Pro $79, and Braintrust Pro $249. None of these list a per-seat charge.

### Can I switch from LangSmith without re-instrumenting my agent?

If your agent emits OpenTelemetry, mostly yes: Langfuse, Arize Phoenix, Opik, Laminar, and Agnost AI all accept OTel traces, so you change the exporter endpoint and headers. If you instrumented with LangSmith's own SDK wrappers or traceable decorator, plan to swap those calls for the new vendor's SDK or an OpenTelemetry instrumentation.
