← All comparisons

Spanlens vs LangSmith · 2026

A 1-line proxy that works with any stack, no SDK wrapping required. MIT and self-hostable today, not a sales call away.

Summary

LangSmith is the right call if you're committed to LangChain or LangGraph. The integration depth is unmatched. Spanlens is the right call if you want a tool that works with anything, installs with a 1-line baseURL swap, ships fully under MIT, and bundles Prompt A/B with statistical testing.

At a glance: Spanlens vs LangSmith (2026)

Side-by-side feature comparison of Spanlens and LangSmith in 2026.
FeatureSpanlensLangSmith
1-line baseURL proxy swapYesNo
Works without any frameworkYesPartial
LangChain & LlamaIndex integrationsYesYes
LangGraph graph topology viewYesYes
Vercel AI SDKYesYes
Per-request log with full bodyYesYes
Cost trackingYesYes
Agent tracing (waterfall)YesYes
Critical Path on agent tracesYesNo
3σ anomaly detectionYesPartial
Versioned prompt libraryYesYes
Public prompt hubNoYes
A/B traffic splitYesYes
Built-in Welch t-test on A/BYesNo
LLM-as-judge scoringYesYes
Human annotation queueYesYes
Judge to human correlation trackingYesPartial
DatasetsYesYes
Model swap recommendations with $ savingsYesNo
Per-model cost breakdownYesYes
Security scanning (API keys, PII, prompt injection)YesPartial
Fully MIT (entire repo)YesNo
Single-command Docker self-hostYesPartial
Managed cloud optionYesYes

Updated 2026-06-03. Scroll for the grouped view with notes below.

Why teams pick Spanlens over LangSmith

Framework-agnostic by design

Spanlens is a proxy. Any HTTP client, any framework, any language sees instant traces. LangSmith markets itself as framework-agnostic too and supports non-LangChain code via SDK wrappers, but its deepest integration (and features like the native graph view) still assume LangChain or LangGraph.

No LangChain lock-in

Adopting LangSmith pushes you toward LangChain abstractions. Spanlens never asks you to rewrite. Keep your raw OpenAI, Anthropic, or Gemini calls, custom orchestration, or whatever framework you already chose.

Fully open source, actually self-hostable

Spanlens is MIT and ships with a single docker-compose self-host. LangSmith Enterprise self-hosting exists but sits behind enterprise sales and pricing.

Drop-in proxy install in 60 seconds

Change one URL. Done. LangSmith requires wrapping your chains, decorating functions, or setting environment variables that only activate when LangChain runs.

Prompt A/B with statistical testing

Spanlens runs prompt variants and reports Welch t-test results on latency and cost, plus a z-test on error rate. LangSmith has experiments but the statistical layer is something you assemble.

Model savings recommender

Spanlens flags routes where a smaller model would match quality and quotes the monthly savings. LangSmith focuses on evals, and cost-tier suggestions are not its primary surface.

Feature-by-feature

Setup & framework coupling
Feature
Spanlens
LangSmith
1-line baseURL proxy swap
LangSmith captures via wrapping or env vars, and only works fully inside LangChain.
Works without any framework
LangChain & LlamaIndex integrations
LangGraph graph topology view
Both render the node graph. LangSmith goes deeper into state-transition introspection.
Vercel AI SDK
Core observability
Feature
Spanlens
LangSmith
Per-request log with full body
Cost tracking
Agent tracing (waterfall)
Critical Path on agent traces
3σ anomaly detection
Prompts & experiments
Feature
Spanlens
LangSmith
Versioned prompt library
Public prompt hub
A/B traffic split
Built-in Welch t-test on A/B
Eval & quality
Feature
Spanlens
LangSmith
LLM-as-judge scoring
Human annotation queue
Judge to human correlation tracking
Datasets
Cost optimization
Feature
Spanlens
LangSmith
Model swap recommendations with $ savings
Per-model cost breakdown
Security
Feature
Spanlens
LangSmith
Security scanning (API keys, PII, prompt injection)
License & deployment
Feature
Spanlens
LangSmith
Fully MIT (entire repo)
Single-command Docker self-host
LangSmith self-host is gated behind enterprise.
Managed cloud option

Last updated 2026-06-03 · Spot something inaccurate? Let us know.

When LangSmith might be the better fit

We don't think every team should pick us. Here's where LangSmith legitimately wins.

You're all-in on LangChain or LangGraph

LangSmith is the most deeply integrated tool for the LangChain stack. Auto-instrumentation of chains, graphs, and tools just works. If you're committed to that ecosystem, LangSmith is the natural choice.

You need first-party LangGraph trace support

LangGraph nodes, edges, and state transitions render natively in LangSmith. Spanlens also renders a graph topology view of your LangGraph runs (with the critical path highlighted), but the depth of state-transition introspection in LangSmith is still ahead. If you need to debug LangGraph state mutations specifically, LangSmith goes deeper.

You want one vendor for framework + observability

Buying LangChain plus LangSmith from the same vendor means one support contract and aligned roadmaps. Spanlens is a separate vendor purposely.

Hub for sharing community prompts

LangSmith Hub has a sizable community of shared prompts and chains, useful for browsing patterns. Spanlens treats prompts as part of your private library; if hub-style discovery is core to your workflow, LangSmith wins on that surface.

Frequently asked questions

Why pick Spanlens over LangSmith for "Framework-agnostic by design"?

Spanlens is a proxy. Any HTTP client, any framework, any language sees instant traces. LangSmith markets itself as framework-agnostic too and supports non-LangChain code via SDK wrappers, but its deepest integration (and features like the native graph view) still assume LangChain or LangGraph.

Why pick Spanlens over LangSmith for "No LangChain lock-in"?

Adopting LangSmith pushes you toward LangChain abstractions. Spanlens never asks you to rewrite. Keep your raw OpenAI, Anthropic, or Gemini calls, custom orchestration, or whatever framework you already chose.

Why pick Spanlens over LangSmith for "Fully open source, actually self-hostable"?

Spanlens is MIT and ships with a single docker-compose self-host. LangSmith Enterprise self-hosting exists but sits behind enterprise sales and pricing.

Why pick Spanlens over LangSmith for "Drop-in proxy install in 60 seconds"?

Change one URL. Done. LangSmith requires wrapping your chains, decorating functions, or setting environment variables that only activate when LangChain runs.

Why pick Spanlens over LangSmith for "Prompt A/B with statistical testing"?

Spanlens runs prompt variants and reports Welch t-test results on latency and cost, plus a z-test on error rate. LangSmith has experiments but the statistical layer is something you assemble.

Why pick Spanlens over LangSmith for "Model savings recommender"?

Spanlens flags routes where a smaller model would match quality and quotes the monthly savings. LangSmith focuses on evals, and cost-tier suggestions are not its primary surface.

When is LangSmith a better fit than Spanlens for "You're all-in on LangChain or LangGraph"?

LangSmith is the most deeply integrated tool for the LangChain stack. Auto-instrumentation of chains, graphs, and tools just works. If you're committed to that ecosystem, LangSmith is the natural choice.

When is LangSmith a better fit than Spanlens for "You need first-party LangGraph trace support"?

LangGraph nodes, edges, and state transitions render natively in LangSmith. Spanlens also renders a graph topology view of your LangGraph runs (with the critical path highlighted), but the depth of state-transition introspection in LangSmith is still ahead. If you need to debug LangGraph state mutations specifically, LangSmith goes deeper.

When is LangSmith a better fit than Spanlens for "You want one vendor for framework + observability"?

Buying LangChain plus LangSmith from the same vendor means one support contract and aligned roadmaps. Spanlens is a separate vendor purposely.

When is LangSmith a better fit than Spanlens for "Hub for sharing community prompts"?

LangSmith Hub has a sizable community of shared prompts and chains, useful for browsing patterns. Spanlens treats prompts as part of your private library; if hub-style discovery is core to your workflow, LangSmith wins on that surface.

If your codebase already breathes LangChain, LangSmith is the safe pick. If you want zero lock-in and a 60-second install, try Spanlens.

Free tier · No credit card · Self-host with Docker