Request early access

Alternatives · LangSmith

6 best LangSmith alternatives in 2026

LangSmith is LangChain's platform for tracing, evaluating, and deploying agents. Its own pricing and docs show why teams shop around. Plus costs $39 per seat a month, and invited users count as seats. Base traces keep 14 days. Self-hosting is an Enterprise add-on. Engine, its automatic issue finder, needs a paid plan and bills about 7 to 45 usage units per run. three.dev needs no setup and has unlimited seats on every plan. Here is what each alternative is best for, what it costs, and where each one beats LangSmith.

Bryan Lee, Product LeadUpdated · 16 min read

Quick comparison: the best LangSmith alternatives

ToolBest forStarting priceKey difference from LangSmithReview
01three.devTeams with a live AI feature and a coding agent, and no time to write evalsFree; paid from $79 a monthFinds and ranks failures by severity on every plan, Free included. Unlimited seats and nothing to turn on.Read the three.dev review
02LangfuseLangSmith users who want to self-host for free or keep traces longer than 14 daysFree; Core $29 a monthOpen source (MIT core) and free to self-host. Paid cloud plans keep 90 days to 3 years of data.Read the Langfuse review
03BraintrustTeams that want evals in CI to decide what ships and want traffic sorted by intent and sentimentFree; Pro $249 a monthEvals first, with CI checks on every pull request. Topics needs an owner to turn it on in each project.Read the Braintrust review
04RaindropTeams that care more about agent failures in production than about evals and deploymentFree; Pro $299 a month plus usageSentry-style monitoring. Pro keeps data with no retention limit.Read the Raindrop review
05Arize PhoenixLangSmith users who want to self-host for free on OpenTelemetryPhoenix free to self-host; Arize AX Pro $50 a monthFree to self-host on OpenTelemetry. Hosted Arize AX adds Signal, capped below Enterprise.Read the Arize Phoenix review
06OpikTeams that want Apache 2.0 open source and a $19 cloud planFree; Pro $19 a monthApache 2.0 and a $19 cloud plan. Diagnostics runs on Ollie credits, on request or daily.Read the Opik review
··LangSmithTeams on LangChain or LangGraph that want tracing, evals, and deployment from one vendorFree for 1 seat; Plus $39 per seat a monthThe tool you are comparing against.
Best for teams with no eval setup yet

1. three.dev

three.dev scores a sample of your live LLM traffic in the background and groups the failures into named failure modes, ranked by severity. LangSmith's Engine finds issues too, on Plus and Enterprise, once it is turned on for the organization and for each tracing project, and it bills usage units per run. three.dev does this on every plan, Free included, with nothing to switch on and no seats to count. Codex, Claude Code, and Cursor then read each failure mode over the three.dev MCP server and propose the fix.

  • Chat

    Ask about your traffic in plain words. Answers link to the requests behind them.

    In Chat, a person clicks Show me how my agent is failing. The answer names Wrong refund window, 386 of 8,124 scored requests, and links one request.

Where it is strong

  • Failure modes on every plan, Free included. No organization switch, project switch, or agent overview to accept first.
  • Scoring comes with the plan, so there are no usage units per scan.
  • The MCP server works with Codex as well as Claude Code and Cursor, and a Fix with your agent button hands one failure mode to the agent.
  • three.dev sets each severity, and your team can change it. Your team can also write a failure mode in plain language, split or merge failure modes, or dismiss one.

Where it falls short

  • No evals, prompt hub, CI checks, alerts, or deployment.
  • three.dev drafts no pull requests itself. Your coding agent writes the change.
  • Requests pass through a proxy at gate.three.dev, where the LangSmith SDK sends traces in the background. There is no self-hosted version, and it is not open source.
  • Tool runs and retrieval steps get no spans of their own; three.dev records the LLM calls and the tool calls the model asks for. Not for health data (PHI) or government workloads yet.
Pricing
Free is $0 with 1 GB of data (about 50,000 requests) and 5k scored requests a month. Starter is $79 a month with 5 GB and 10k scored requests, then $3 per GB and $2.50 per 1k scored. Pro is $249 a month with 20 GB and 25k scored requests, then $2 per GB and $1.50 per 1k scored. Every plan includes scoring and unlimited seats.
Choose it over LangSmith if
you want failures found and ranked with no seats to count and no usage units per scan, and your coding agent makes the fix.
Request early accessChat with usSee pricing

Why teams choose three.dev

"Using three.dev has been a huge leap for us in providing our users the highest quality while maintaining fast responses."

Don PottingerCo-founder and CTO, LanguaTalkRead the case study

three.dev vs LangSmith

Featurethree.devLangSmith
What it isFailure-mode discovery for live AI features, with the fix handed to your coding agentLangChain's agent engineering platform: tracing, evals, prompts, deployment, and Engine
How you connectPoint your model client, LangChain included, at gate.three.dev with a three.dev API key and one header.Add a LangSmith SDK (Python, TypeScript, Go, or Java), native tracing for popular agent frameworks, or OpenTelemetry.
Setup before you see failuresNone. Scoring starts on your live traffic.For Engine: a Plus or Enterprise plan, an admin turns it on for the organization, someone turns it on for each tracing project, and you accept the agent overview it writes. First results take up to 20 minutes.
Grouping failuresAutomatic, on every plan. Issues group into named failure modes, each with a severity, most severe first.Engine scans on a schedule, groups failures into issues, and ranks them by severity. Insights reports cluster up to 1,000 traces each.
Shaping the listYour team can change any severity three.dev sets, write a new failure mode in plain language, and split, merge, or dismiss failure modes.You choose focus areas and an analysis level when you turn Engine on. Engine can file an issue from a single trace and reopens an issue that comes back.
Fixing themThe MCP server hands your coding agent a named failure mode, its failing requests, and why each one failed. Your agent proposes the change.Engine drafts prompt and code fixes, opens GitHub pull requests or GitLab merge requests, and can hand an issue to a coding agent.
Coding agents supportedClaude Code and Codex plugins set up the MCP server. Cursor and other clients take the URL.A Remote MCP server for Claude Code and Cursor. Its docs say the Codex CLI does not work with it and point to the LangSmith CLI.
Checking a fixTag requests with a release, then compare two releases or two date ranges on quality, latency, and cost.Offline evals on datasets, and fix validation (private beta, for agents on LangSmith Cloud deployments).
Evals, prompts, deploymentNot offered todayOnline and offline evals, pytest and Vitest CI checks, a prompt hub, and Deployment
AlertsNot offered todayThreshold alerts to Slack, PagerDuty, Dynatrace, or a webhook
SeatsUnlimited on every plan$39 per seat a month on Plus, invited users included
History kept30 days on Free, 90 on Starter, 180 on Pro14 days for base traces; 180 days for extended traces, at an extra charge
HostingHosted proxy at gate.three.devCloud; self-hosted and hybrid on Enterprise
PriceFree plan; Starter $79, Pro $249 a month. Scoring included. Unlimited users.Developer free for 1 seat; Plus $39 per seat a month plus usage; Engine billed in $1 units; Enterprise custom

What you gain moving from LangSmith

  • Failure modes found on every plan, Free included, with no switch to turn on
  • Unlimited seats, so the price does not grow with your team
  • Scoring included in the plan price, with no usage units per scan
  • Severities your team can change, and failure modes it can create, split, merge, or dismiss
  • One MCP server that works with Codex, Claude Code, and Cursor

What you give up

  • Engine's drafted fix pull requests and merge requests
  • Evals, prompts, and CI checks
  • Alerts and deployment
  • Full traces of tool runs and retrieval steps
Best open-source pick for tracing, evals, and prompts

2. Langfuse

Langfuse covers the same core as LangSmith, with tracing, evals, prompt management, datasets, and experiments, and it is open source. Its core is MIT licensed and free to self-host, where LangSmith self-hosts only as an Enterprise add-on. ClickHouse acquired Langfuse in January 2026 and says there are no planned licensing changes. Langfuse has no counterpart to Engine yet. Its grouping feature, Topics, is "In progress" on the public roadmap and in a closed beta on Langfuse Cloud in October 2026, so today you score traffic with judge prompts you write and cluster the errors yourself.

Where it is strong

  • The MIT core self-hosts for free with no usage cap, and the free build includes Enterprise SSO and RBAC.
  • Longer history on paid plans: Core keeps 90 days and Pro 3 years, against 14 days for LangSmith base traces.
  • Its MCP server can write as well as read, and a CLI wraps the whole API, so an agent can edit prompts, datasets, and scores.
  • Prompts, datasets, experiments, and annotation queues on every plan, plus a GitHub Action that fails a run when an experiment score drops below your threshold.

Where it falls short

  • No generally available automatic failure grouping. Topics is in closed beta for allowlisted projects on Langfuse Cloud.
  • Scoring live traffic is setup you own: an LLM connection, a judge prompt, and a rule with filters and a sampling rate. Judge calls bill to your model key, and each score is a billable unit.
  • Evaluators read one observation at a time. Session and multi-span evaluations are on the roadmap.
  • Self-hosting means running Postgres, ClickHouse, Redis or Valkey, and S3-type storage.
Pricing
Hobby is free with 50k units a month, 30 days of data, and 2 users. Core is $29 a month and Pro $199, each with 100k units, then $8 per 100k, falling to $6 above 50M. A unit is any trace, observation, or score. Paid plans have unlimited users. On Cloud, SSO and fine-grained RBAC need the $300-a-month Teams add-on on Pro, or Enterprise at $2,499 a month.
Choose it over LangSmith if
you want open source you can self-host for free and a longer trace history, and you are ready to write your own judges.
Visit Langfuse(opens in a new tab)
Best for evals and CI checks on every pull request

3. Braintrust

Braintrust is an eval and observability platform: tracing, datasets, scorers, experiments, CI checks, prompts, and alerts. It works with LangChain and LangGraph, so you can keep your framework. Where LangSmith has Insights reports, Braintrust has Topics, which clusters production traces by task, sentiment, issues, and any custom facets you add. Patterns, in public preview, looks for recurring problems on a schedule and ranks them by severity and frequency.

Where it is strong

  • Topics clusters every trace by what the user asked for, how they felt, and what went wrong, with built-in facets plus any custom facets you write.
  • Loop drafts scorers, prompts, datasets, and dashboards with your approval, and can open GitHub pull requests (public preview).
  • SDKs in six languages, where LangSmith has four, plus OpenTelemetry and integrations for LangChain and LangGraph.
  • Evals run in CI on every pull request, with results posted to the pull request and an option to gate merges.

Where it falls short

  • Not published as open source. Self-hosting and BYOC are Enterprise only, with Braintrust running the control plane.
  • Nothing between free Starter and Pro at $249 a month.
  • Topics needs a member of the Owners group to turn it on in each project, and Patterns is still in public preview.
  • Topics and Loop use monthly model credits that do not roll over. Starter pauses Topics when its credit runs out unless a card is on file.
Pricing
Braintrust meters processed data, scores, and model credits. Like three.dev, every plan has unlimited users. Starter is free: 1 GB, 10k scores, and $10 of credits a month. Pro is $249 a month: 5 GB, 50k scores, $100 of credits, and 30-day logs, extendable at $0.50 per GB a month. Enterprise is custom.
Choose it over LangSmith if
you want evals and CI checks to decide what ships, and you want production traffic sorted by intent and sentiment.
Visit Braintrust(opens in a new tab)
Best for alerts on agents in production

4. Raindrop

Raindrop is a monitoring platform for AI agents that works the way Sentry does for web apps: tracing, search, signals, alerts, and experiments. It is narrower than LangSmith and aimed at production. Signals track behaviors you describe in plain language, and Signal Alerts post to Slack when one crosses a threshold. On Pro, Issue Detection looks for new breakage patterns and regressions after a deploy, and gives each issue a severity and a trend.

Where it is strong

  • Pro keeps data with no retention limit, against 14 days for LangSmith base traces.
  • Issue Detection watches for regressions after a deploy and failures spreading across users, and tracks each issue's trend.
  • Slack alerts fire in real time when a signal crosses a threshold.
  • SDKs for TypeScript, Python, and Go, with Rust, Swift, and Java in beta, plus OpenTelemetry over OTLP/HTTP.

Where it falls short

  • Issue Detection, Stumbles, and Experiments need Pro: $299 a month plus $0.003 per event up to 1M.
  • Hobby stops recording for the rest of the month after 1,000 events.
  • Custom signals are capped at 1 on Hobby and 10 on Pro.
  • Hosted, with self-hosting in your own cloud in beta.
Pricing
Hobby is free: 1,000 events, 14 days of retention, and 1 custom signal a month. Pro is $299 a month plus $0.003 per event up to 1M, then $0.002, with unlimited retention, 10 custom signals, and a 14-day free trial. One event is a user message and the agent's response, including tool calls and sub-agents, or one background agent run. Enterprise adds SSO, audit logs, and warehouse exports.
Choose it over LangSmith if
you want Slack alerts and regression watch on production agents more than evals and deployment, and a $299 monthly base fits.
Visit Raindrop(opens in a new tab)
Best open-source pick on OpenTelemetry

5. Arize Phoenix

Phoenix is Arize's open-source tool for tracing, evals, datasets, experiments, and prompts. It is built on OpenTelemetry and self-hosts for free, which LangSmith offers only as an Enterprise add-on. Arize AX, the hosted platform, adds online evals, alerts, and Signal, which scans traces on a schedule and groups recurring failures into ranked issues with a proposed fix. Like Engine, Signal can open fix pull requests, on Enterprise. Dynatrace completed its purchase of Arize on October 1, 2026.

Where it is strong

  • Phoenix self-hosts for free with no feature gates and can run fully air-gapped.
  • Evals run as pytest, Vitest, or Jest tests that record to Phoenix and gate CI.
  • A Signal issue can become a running evaluator: Alyx builds one for that failure pattern and starts scoring new traces.
  • Built on OpenTelemetry, so the tracing you add is not tied to one vendor's SDK.

Where it falls short

  • Online evals on live traffic, alerting, and Signal live in Arize AX. Phoenix's docs send you there.
  • Phoenix uses the Elastic License 2.0, which bars offering it to others as a hosted service.
  • Signal stops at 10 issues a month on AX Free and 25 on Pro, and fix pull requests are Enterprise only.
  • Dynatrace plans to fold Arize into its own platform over time, while it keeps supporting Phoenix and AX.
Pricing
Phoenix costs nothing to self-host. Arize AX Free: 25k spans, 1 GB a month, 15 days of retention. AX Pro: $50 a month, 50k spans and 10 GB, then $0.0008 per span and $3 per GB, with 30 days of retention. Enterprise: custom, with self-hosted AX.
Choose it over LangSmith if
you want tracing and evals you run yourself on OpenTelemetry, without an Enterprise contract.
Visit Arize Phoenix(opens in a new tab)
Best Apache-licensed pick with failure diagnostics

6. Opik

Opik is Comet's open-source platform for tracing, evals, prompt management, and production monitoring, under the Apache 2.0 license. Its counterpart to Engine is Diagnostics: the Ollie assistant scans your traces on request, or every day once you switch that on, groups repeated failures into one issue each, and suggests a root cause and a fix. Engine opens pull requests. Ollie edits code on your machine through Opik Connect and waits for your approval on each edit.

Where it is strong

  • Pro is $19 a month for 100k spans, the lowest paid plan on this list.
  • Apache 2.0, and the hosted versions run the same codebase as the open-source build.
  • The free self-hosted build has online evaluation, alerts, 30+ built-in metrics, and AI guardrails.
  • Alerts reach Slack, PagerDuty, or any webhook.

Where it falls short

  • Ollie credits pay for Diagnostics, and Comet does not publish their price.
  • Daily scans start only when someone turns them on in a project, cover the previous day, and never go back for missed days.
  • SSO and RBAC are Enterprise only, and self-hosted Opik has no user management.
  • To self-host, you run ClickHouse, MySQL, Redis, and object storage yourself.
Pricing
Self-hosting Opik is free. Opik Cloud Free: 25k spans a month and 60 days of retention. Pro: $19 a month for 100k spans, then $5 per 100k. Enterprise: custom. Diagnostics and Ollie draw on credits on top, at a price Comet does not publish.
Choose it over LangSmith if
you want an Apache 2.0 tool you can self-host, or a hosted plan at $19 a month.
Visit Opik(opens in a new tab)

Which tool is right for you?

  • You want failures found, ranked by severity, and handed to your coding agent, with no per-seat pricethree.dev
  • You want open-source tracing, prompts, and evals you can self-host for freeLangfuse
  • You want evals and CI checks to decide what shipsBraintrust
  • You want Slack alerts and no retention limit for agents in productionRaindrop
  • You want OpenTelemetry-native tracing and evals you run yourself, without an Enterprise contractArize Phoenix
  • You want an Apache 2.0 tool with a $19 cloud planOpik

Questions

What is the best LangSmith alternative?

That depends on what pushes you off LangSmith. If it is the per-seat bill or Engine's usage units, three.dev finds and ranks failures on every plan with unlimited seats. If it is closed source, Langfuse, Arize Phoenix, and Opik self-host for free. If evals in CI are the center of your work, Braintrust. If you want Slack alerts on production agents, Raindrop.

Which LangSmith alternatives find production failures automatically?

three.dev needs no setup: it scores a sample of your requests in the background, reports each defect as an issue, and groups issues into named failure modes, ranked by severity, once enough collect. Every other tool here names a condition in its own docs (checked October 29, 2026), as Engine does with its paid plan, admin switch, and usage units. Braintrust: an Owners-group member turns Topics on in each project; clusters appear after 100 facet summaries, refresh daily, and draw on model credits that do not roll over. Patterns, in public preview, is a daily Loop run that someone turns on per project. Raindrop: Issue Detection comes with Pro, $299 a month plus $0.003 per event. Arize: AX Signal is switched on per project, runs every 6 hours by default, and caps at 10 issues a month on Free and 25 on Pro; Phoenix does not include it. Opik: Diagnostics runs on Ollie credits at an unpublished price, and daily scans start only when someone switches them on. Langfuse: Topics is "In progress" on the public roadmap and, per its GitHub release notes, in a closed beta on Langfuse Cloud in October 2026.

Is there a LangSmith alternative without per-seat pricing?

Yes. three.dev has unlimited seats on every plan, Free included, with paid plans from $79 a month. Braintrust has unlimited users on every plan, with Pro at $249 a month. Langfuse has unlimited users on its paid cloud plans from $29 a month (its free Hobby plan has 2), and self-hosting it is free. LangSmith Plus is $39 per seat a month, and it counts invited users as seats.

Which LangSmith alternatives can I self-host?

Langfuse (MIT core), Arize Phoenix (Elastic License 2.0), and Opik (Apache 2.0) are free to self-host. Braintrust self-hosts on Enterprise only, and Braintrust still runs the control plane. Raindrop lists self-hosting in your own cloud as beta. three.dev is hosted only. LangSmith itself self-hosts as an add-on to its Enterprise plan.

LangSmith vs Langfuse: which is better?

LangSmith fits teams on LangChain and LangGraph, and its Engine finds issues and drafts fix pull requests on Plus and Enterprise, billed per run. It charges $39 per seat on Plus, keeps base traces 14 days, and self-hosts only on Enterprise. Langfuse's core is MIT licensed and free to self-host, its paid cloud plans start at $29 a month and keep 90 days or more, and its automatic grouping (Topics) is in closed beta. Pick LangSmith for LangChain depth and Engine. Pick Langfuse for open source and free self-hosting.

LangSmith vs Braintrust: which is better?

LangSmith charges $39 per seat on Plus and counts invited users. Braintrust, like three.dev, has unlimited users on every plan, with Pro at $249 a month. Both find problems on their own: LangSmith Engine on Plus and Enterprise, billed per run, and Braintrust Topics once an owner turns it on in each project, with Patterns in public preview. Engine drafts fix pull requests on GitHub or GitLab. Braintrust's Loop can open GitHub pull requests in preview, and its GitHub Action runs evals on each pull request. Both self-host only on Enterprise. Pick LangSmith for LangChain and LangGraph apps. Pick Braintrust if evals in CI decide what ships.

Can I keep LangChain and leave LangSmith?

Yes. three.dev documents LangChain setup: you point the model client at gate.three.dev, add a three.dev API key, and add the X-Three-Use-Case header. Braintrust documents LangChain and LangGraph integrations, Langfuse lists 100+ integrations plus OpenTelemetry, and Arize Phoenix is built on OpenTelemetry.

Can I use three.dev and LangSmith together?

Yes. three.dev only changes the base URL your model client calls, so LangSmith tracing can stay in place. You can keep LangSmith for evals and deployment and add three.dev to find and rank your production failures.

How do I move from LangSmith to three.dev?

Paste this prompt into your coding agent: "Install github.com/round3ai/skills and run its three-dev-setup skill in this repository." It changes the base URL to gate.three.dev, adds a three.dev API key, and adds the X-Three-Use-Case header. Scoring then starts on your live traffic.

LangSmith sends traces in the background. Does three.dev slow down my requests?

three.dev is a proxy at gate.three.dev, so your requests pass through it. Scoring runs after the response returns, so it never changes a request or adds latency.

How we made this list

We make three.dev, so we list it first. For LangSmith, we read LangChain's pricing page, the Engine, Insights, billing, and self-hosting docs, and the LangSmith changelog on October 29, 2026. We read every other tool's own site, docs, and pricing page the same day and wrote each review for a team leaving LangSmith, including where that tool is stronger than three.dev. Check each pricing page before you buy; prices and features change often.

Sources

Ship on evidence, not vibes.

See how your agent fails. Hand the fix to your coding agent.

Request early accessChat with us