Cycling through all six. Tap any point to stop.

Measured on six things

Open-source LLM observability – one line of code to monit...

Growth

Very new (June 2026 launch), active GitHub with frequent commits, but zero external reviews or community traction yet.

Screenshots

1 / 1

Spanlens

About this app

I've been testing Spanlens for the last few days, and I'm genuinely impressed. It's an open-source (MIT) LLM observability platform that logs every request to OpenAI, Anthropic, and Gemini with just one line of code. Setup is ridiculously easy – you simply swap your provider SDK for Spanlens's drop-in, and boom, you get request logging, cost tracking, agent tracing, anomaly detection, PII scanning, model recommendations, evals, and experiments. The self-hosted option means your data never leaves your VPC, which is huge for compliance. I love the waterfall traces for debugging multi-step agents – it shows critical path, cost attribution, and latency outliers instantly. The cost dashboard is a lifesaver for understanding why your AI bill spiked. And the eval/experiment workflow lets you replay real traffic against different models and prompt versions, so you can actually measure quality improvements before shipping. With a free tier for 50K requests/month and Pro at $29/month, it's in...

Key features

1One-line drop-in SDK for OpenAI/Anthropic/Gemini with full request logging and cost tracking.
2Waterfall agent tracing with critical path, cost attribution, and anomaly detection.
3Evals, experiments, model recommendations, PII scanning, and self-hosting via Docker.

Target customer profile

Drawn from the product itself, not from a survey.

Demographic

AI/ML engineers building LLM-powered applications

Pain points

Debugging black-box LLM calls, tracking costs across multiple providers, identifying latency bottlenecks

Primary needs

One-line integration, detailed request logs, cost breakdowns, agent tracing

Demographic

Platform teams monitoring production AI workloads

Pain points

Anomaly detection, PII leakage, compliance requirements, multi-team budget management

Primary needs

Security scanning, anomaly alerts, role-based access, audit logs, self-hosting

Demographic

Product managers optimizing AI feature quality and cost

Pain points

Comparing model performance, validating prompt improvements, controlling AI spending without sacrificing quality

Primary needs

Eval scores, experiment comparisons, model cost recommendations, dataset replay

Scout Review

Written by AI from measured evidence, scored out of 100.

Spanlens is a fresh, MIT-licensed LLM observability tool that excels at dead-simple integration and practical features like agent tracing, cost recommendations, and experiment replay. It directly addresses pain points in the provided alternatives—easier setup than Langfuse, more evals than Helicone, and self-hosting unlike LangSmith. The free tier and flat pricing are attractive, but its youth means real-world validation is still pending.

Measure by measure

Usability

Extremely easy one-line/CLI setup, rich docs, live demo, and framework integrations make onboarding fast for LLM devs.

Design

Polished, modern landing with live interactive demo, clean visuals for traces/costs, and consistent branding across sections.

Speed

Site and demo feel snappy; vendor claims sub-3ms overhead with async design, though no independent benchmarks exist.

Security

Strong self-host + PII scanning story; MIT license and Docker deployment, but no audits or compliance certifications published.

Accessibility

Responsive modern site with readable demo tables; no WCAG claims, alt text details, or accessibility documentation found.

Growth

Very new (June 2026 launch), active GitHub with frequent commits, but zero external reviews or community traction yet.

Improvement areas

Lack of third-party security audits or compliance certifications.
Zero independent user reviews or community traction metrics.
No published accessibility or performance benchmark data.

Design & Usability

Spanlens delivers a genuinely polished landing experience with an interactive live demo that showcases request logs, cost tracking, and waterfall traces in real time. The one-line SDK swap and CLI wizard make it feel instantly usable for AI engineers already working with OpenAI/Anthropic/Gemini. Documentation is thorough, covering self-hosting, LangChain callbacks, and experiments. Compared to peers like Langfuse (more complex SDK) or Helicone (lighter on evals), the onboarding friction is noticeably lower. The UI emphasizes practical dashboards over flashy animations, which fits the target audience of platform teams and PMs who need cost visibility and agent debugging fast.

Speed & Security

Performance claims center on sub-3ms overhead via async ingestion, and the demo dashboard renders thousands of events without lag. Self-hosting via a single Docker command keeps data in your VPC, paired with built-in PII regex scanning and secret masking—strong for compliance-sensitive workloads. However, the absence of published SOC 2, ISO, or independent audits keeps the security posture in the 'solid but unremarkable' band. No breach history exists simply because the product is brand new. Speed and security are competent table stakes rather than differentiators versus established alternatives.

Accessibility & Growth

The site is responsive and the demo tables are readable, but there are no accessibility statements, WCAG mentions, or multi-language support. Growth is the weakest dimension: launched around June 2026, the GitHub repo shows healthy commit activity and a clear positioning table versus Langfuse/Helicone, yet independent reviews, Product Hunt traction, or community size are nonexistent. With only demo data claiming 1,204 users and no G2/Capterra presence, Spanlens is still in the 'promising early project' phase rather than proven product.

Conclusion

If you're an AI engineer tired of black-box LLM spend or debugging multi-step agents, Spanlens is worth a 30-second test drive via the live demo. Just don't bet production compliance on it until audits appear.

Suggestions for improvement

Publish SOC 2 or equivalent audit to boost enterprise trust.
Add explicit WCAG 2.1 AA compliance notes and test results.
Seed G2/Capterra listings and collect early user reviews.
Publish independent latency benchmarks versus Langfuse/Helicone.

How we score

Alternatives comparison

Named competitors, point by point. Nobody paid to appear here or to be left out.

	Spanlens	Langfuse	Helicone	LangSmith	Arize Phoenix
Integration complexity	One-line SDK swap or CLI init	More complex SDK setup	Proxy-based, moderate complexity	LangChain-native, framework lock-in	More complex setup for evals
Self-hosting	Docker one-liner, free forever	Supported, more involved	Supported	Not self-hostable	Open-source, self-hostable
Evals & experiments	LLM-as-judge + dataset replay	Strong evals & prompt management	Limited evals focus	LangChain-centric evals	Strong LLM evals & drift detection