A 1-line proxy that works with any stack, no SDK wrapping required. MIT and self-hostable today, not a sales call away.
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.
| Feature | Spanlens | LangSmith |
|---|---|---|
| 1-line baseURL proxy swap | Yes | No |
| Works without any framework | Yes | Partial |
| LangChain & LlamaIndex integrations | Yes | Yes |
| LangGraph graph topology view | Yes | Yes |
| Vercel AI SDK | Yes | Yes |
| Per-request log with full body | Yes | Yes |
| Cost tracking | Yes | Yes |
| Agent tracing (waterfall) | Yes | Yes |
| Critical Path on agent traces | Yes | No |
| 3σ anomaly detection | Yes | Partial |
| Versioned prompt library | Yes | Yes |
| Public prompt hub | No | Yes |
| A/B traffic split | Yes | Yes |
| Built-in Welch t-test on A/B | Yes | No |
| LLM-as-judge scoring | Yes | Yes |
| Human annotation queue | Yes | Yes |
| Judge to human correlation tracking | Yes | Partial |
| Datasets | Yes | Yes |
| Model swap recommendations with $ savings | Yes | No |
| Per-model cost breakdown | Yes | Yes |
| Security scanning (API keys, PII, prompt injection) | Yes | Partial |
| Fully MIT (entire repo) | Yes | No |
| Single-command Docker self-host | Yes | Partial |
| Managed cloud option | Yes | Yes |
Updated 2026-06-03. Scroll for the grouped view with notes below.
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.
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.
Spanlens is MIT and ships with a single docker-compose self-host. LangSmith Enterprise self-hosting exists but sits behind enterprise sales and pricing.
Change one URL. Done. LangSmith requires wrapping your chains, decorating functions, or setting environment variables that only activate when LangChain runs.
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.
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.
Last updated 2026-06-03 · Spot something inaccurate? Let us know.
We don't think every team should pick us. Here's where LangSmith legitimately wins.
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.
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.
Buying LangChain plus LangSmith from the same vendor means one support contract and aligned roadmaps. Spanlens is a separate vendor purposely.
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.
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.
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.
Spanlens is MIT and ships with a single docker-compose self-host. LangSmith Enterprise self-hosting exists but sits behind enterprise sales and pricing.
Change one URL. Done. LangSmith requires wrapping your chains, decorating functions, or setting environment variables that only activate when LangChain runs.
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.
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.
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.
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.
Buying LangChain plus LangSmith from the same vendor means one support contract and aligned roadmaps. Spanlens is a separate vendor purposely.
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